image — machine vision

The image module is used for machine vision.

Functions

image.binary_to_grayscale

image.binary_to_grayscale(binary_image_value: 0 | 1) -> int

Returns a converted binary value (0-1) to a grayscale value (0-255).

image.binary_to_rgb

image.binary_to_rgb(binary_image_value: 0 | 1) -> Tuple[int, int, int]

Returns a converted binary value (0-1) to a 3 value RGB888 tuple.

image.binary_to_lab

image.binary_to_lab(binary_image_value: 0 | 1) -> Tuple[int, int, int]

Returns a converted binary value (0-1) to a 3 value LAB tuple.

L goes between 0 and 100 and A/B go from -128 to 128.

image.binary_to_yuv

image.binary_to_yuv(binary_image_value: 0 | 1) -> Tuple[int, int, int]

Returns a converted binary value (0-1) to a 3 value YUV tuple.

Y goes between 0 and 255 and U/V go from -128 to 128.

image.grayscale_to_binary

image.grayscale_to_binary(grayscale_value: int) -> 0 | 1

Returns a converted grayscale value (0-255) to a binary value (0-1).

image.grayscale_to_rgb

image.grayscale_to_rgb(grayscale_value: int) -> Tuple[int, int, int]

Returns a converted grayscale value to a 3 value RGB888 tuple.

Note

The OpenMV Cam firmware does the conversion using a RGB565->RGB888 process so this method won’t return the exact values as a pure RGB888 system would. However, it’s true to how the image lib works internally.

image.grayscale_to_lab

image.grayscale_to_lab(grayscale_value: int) -> Tuple[int, int, int]

Returns a converted grayscale value to a 3 value LAB tuple.

L goes between 0 and 100 and A/B go from -128 to 128.

Note

The OpenMV Cam firmware does the conversion using a RGB565->LAB process so this method won’t return the exact values as a pure LAB system would. However, it’s true to how the image lib works internally.

image.grayscale_to_yuv

image.grayscale_to_yuv(grayscale_value: int) -> Tuple[int, int, int]

Returns a converted grayscale value to a 3 value YUV tuple.

Y goes between 0 and 255 and U/V go from -128 to 128.

Note

The OpenMV Cam firmware does the conversion using a RGB565->YUV process so this method won’t return the exact values as a pure YUV system would. However, it’s true to how the image lib works internally.

image.rgb_to_binary

image.rgb_to_binary(rgb_tuple: Tuple[int, int, int]) -> 0 | 1

Returns a converted 3 value RGB888 tuple to a center range thresholded binary value (0-1).

Note

The OpenMV Cam firmware does the conversion using a RGB888->RGB565 process so this method won’t return the exact values as a pure RGB888 system would. However, it’s true to how the image lib works internally.

image.rgb_to_grayscale

image.rgb_to_grayscale(rgb_tuple: Tuple[int, int, int]) -> int

Returns a converted 3 value RGB888 tuple to a grayscale value (0-255).

Note

image.rgb_to_lab

image.rgb_to_lab(rgb_tuple: Tuple[int, int, int]) -> Tuple[int, int, int]

Returns a converted 3 value RGB888 tuple to a 3 value LAB tuple.

L goes between 0 and 100 and A/B go from -128 to 128.

Note

image.rgb_to_yuv

image.rgb_to_yuv(rgb_tuple: Tuple[int, int, int]) -> Tuple[int, int, int]

Returns a converted 3 value RGB888 tuple to a 3 value YUV tuple.

Y goes between 0 and 255 and U/V go from -128 to 128.

Note

image.lab_to_binary

image.lab_to_binary(lab_tuple: Tuple[int, int, int]) -> 0 | 1

Returns a converted 3 value LAB tuple to a center range thresholded binary value (0-1).

Note

The OpenMV Cam firmware does the conversion using a LAB->RGB565 process so this method won’t return the exact values as a pure LAB system would. However, it’s true to how the image lib works internally.

image.lab_to_grayscale

image.lab_to_grayscale(lab_tuple: Tuple[int, int, int]) -> int

Returns a converted 3 value LAB tuple to a grayscale value (0-255).

Note

image.lab_to_rgb

image.lab_to_rgb(lab_tuple: Tuple[int, int, int]) -> Tuple[int, int, int]

Returns a converted 3 value LAB tuple to a 3 value RGB888 tuple.

Note

image.lab_to_yuv

image.lab_to_yuv(lab_tuple: Tuple[int, int, int]) -> Tuple[int, int, int]

Returns a converted 3 value LAB tuple to a 3 value YUV tuple.

Y goes between 0 and 255 and U/V go from -128 to 128.

Note

image.yuv_to_binary

image.yuv_to_binary(yuv_tuple: Tuple[int, int, int]) -> 0 | 1

Returns a converted 3 value YUV tuple to a center range thresholded binary value (0-1).

Note

The OpenMV Cam firmware does the conversion using a YUV->RGB565 process so this method won’t return the exact values as a pure YUV system would. However, it’s true to how the image lib works internally.

image.yuv_to_grayscale

image.yuv_to_grayscale(yuv_tuple: Tuple[int, int, int]) -> int

Returns a converted 3 value YUV tuple to a grayscale value (0-255).

Note

image.yuv_to_rgb

image.yuv_to_rgb(lab_tuple: Tuple[int, int, int]) -> Tuple[int, int, int]

Returns a converted 3 value YUV tuple to a 3 value RGB888 tuple.

Note

image.yuv_to_lab

image.yuv_to_lab(yuv_tuple: Tuple[int, int, int]) -> Tuple[int, int, int]

Returns a converted 3 value YUV tuple to a 3 value LAB tuple.

L goes between 0 and 100 and A/B go from -128 to 128.

Note

image.load_decriptor

image.load_decriptor(path: str)

Loads a descriptor object from disk.

path is the path to the descriptor file to load.

image.save_descriptor

image.save_descriptor(path: str, descriptor)

Saves the descriptor object descriptor to disk.

path is the path to the descriptor file to save.

image.match_descriptor

image.match_descriptor(descritor0, descriptor1, threshold=70, filter_outliers=False)

For LBP descriptors this function returns an integer representing the difference between the two descriptors. You may then threshold/compare this distance metric as necessary. The distance is a measure of similarity. The closer it is to zero the better the LBP keypoint match.

For ORB descriptors this function returns the kptmatch object. See above.

threshold is used for ORB keypoints to filter ambiguous matches. A lower threshold value tightens the keypoint matching algorithm. threshold may be between 0-100 (int). Defaults to 70.

filter_outliers is used for ORB keypoints to filter out outlier keypoints allow you to raise the threshold. Defaults to False.

class HaarCascade – Feature Descriptor

The Haar Cascade feature descriptor is used for the Image.find_features() method. It doesn’t have any methods itself for you to call.

image.HaarCascade

class image.HaarCascade(path: str, stages: int | None = None)

Loads a Haar Cascade into memory from a Haar Cascade binary file formatted for your OpenMV Cam. If you pass “frontalface” instead of a path then this constructor will load the built-in frontal face Haar Cascade into memory. Additionally, you can also pass “eye” to load a Haar Cascade for eyes into memory. Finally, this method returns the loaded Haar Cascade object for use with Image.find_features().

stages defaults to the number of stages in the Haar Cascade. However, you can specify a lower number of stages to speed up processing the feature detector at the cost of a higher rate of false positives.

Note

You can make your own Haar Cascades to use with your OpenMV Cam. First, Google for “<thing> Haar Cascade” to see if someone already made an OpenCV Haar Cascade for an object you want to detect. If not… then you’ll have to generate your own (which is a lot of work). See here for how to make your own Haar Cascade. Then see this script for converting OpenCV Haar Cascades into a format your OpenMV Cam can read.

Q: What is a Haar Cascade?

A: A Haar Cascade is a series of contrast checks that are used to determine if an object is present in the image. The contrast checks are split of into stages where a stage is only run if previous stages have already passed. The contrast checks are simple things like checking if the center vertical of the image is lighter than the edges. Large area checks are performed first in the earlier stages followed by more numerous and smaller area checks in later stages.

Q: How are Haar Cascades made?

A: Haar Cascades are made by training the generator algorithm against positive and negative labeled images. For example, you’d train the generator algorithm against hundreds of pictures with cats in them that have been labeled as images with cats and against hundreds of images with not cat like things labeled differently. The generator algorithm will then produce a Haar Cascade that detects cats.

class Similarity – Similarity Object

The similarity object is returned by Image.get_similarity().

image.Similarity

class image.Similarity

Please call Image.get_similarity() to create this object.

mean

mean() -> float

Returns the mean of the similarity values computed across the image (float).

You may also get this value doing [0] on the object.

stdev

stdev() -> float

Returns the standard deviation of the similarity values computed across the image ( (float).

You may also get this value doing [1] on the object.

min

min() -> float

Returns the min of the similarity values computed across the image ( (float).

Generally, for the SSIM you want to threshold the min value to determine if two images are different.

You may also get this value doing [2] on the object.

max

max() -> float

Returns the max of the similarity values computed across the image ( (float).

Generally, for the DSIM you want to threshold the max value to determine if two images are different.

You may also get this value doing [3] on the object.

class Histogram – Histogram Object

The histogram object is returned by Image.get_histogram().

Grayscale histograms have one channel with some number of bins. All bins are normalized so that all bins sum to 1.

RGB565 histograms have three channels with some number of bins each. All bins are normalized so that all bins in a channel sum to 1.

image.histogram

class image.histogram

Please call Image.get_histogram() to create this object.

bins

bins() -> List[float]

Returns a list of floats for the grayscale histogram.

You may also get this value doing [0] on the object.

l_bins

l_bins() -> List[float]

Returns a list of floats for the RGB565 histogram LAB L channel.

You may also get this value doing [0] on the object.

a_bins

a_bins() -> List[float]

Returns a list of floats for the RGB565 histogram LAB A channel.

You may also get this value doing [1] on the object.

b_bins

b_bins() -> List[float]

Returns a list of floats for the RGB565 histogram LAB B channel.

You may also get this value doing [2] on the object.

get_percentile

get_percentile(percentile) -> percentile

Computes the CDF of the histogram channels and returns a image.percentile object with the values of the histogram at the passed in percentile (0.0

1.0) (float). So, if you pass in 0.1 this method will tell you (going from left-to-right in the histogram) what bin when summed into an accumulator caused the accumulator to cross 0.1. This is useful to determine min (with 0.1) and max (with 0.9) of a color distribution without outlier effects ruining your results for adaptive color tracking.

get_threshold

get_threshold() -> threshold

Uses Otsu’s Method to compute the optimal threshold values that split the histogram into two halves for each channel of the histogram. This method returns a image.threshold object. This method is particularly useful for determining optimal Image.binary() thresholds.

get_statistics

get_statistics() -> statistics

Computes the mean, median, mode, standard deviation, min, max, lower quartile, and upper quartile of each color channel in the histogram and returns a statistics object.

You may also use histogram.statistics() and histogram.get_stats() as aliases for this method.

class Percentile – Percentile Object

The percentile object is returned by histogram.get_percentile().

Grayscale percentiles have one channel. Use the non l_*, a_*, and b_* method.

RGB565 percentiles have three channels. Use the l_*, a_*, and b_* methods.

image.percentile

class image.percentile

Please call histogram.get_percentile() to create this object.

value

value() -> int

Return the grayscale percentile value (between 0 and 255).

You may also get this value doing [0] on the object.

l_value

l_value() -> int

Return the RGB565 LAB L channel percentile value (between 0 and 100).

You may also get this value doing [0] on the object.

a_value

a_value() -> int

Return the RGB565 LAB A channel percentile value (between -128 and 127).

You may also get this value doing [1] on the object.

b_value

b_value() -> int

Return the RGB565 LAB B channel percentile value (between -128 and 127).

You may also get this value doing [2] on the object.

class Threshold – Threshold Object

The threshold object is returned by histogram.get_threshold().

Grayscale thresholds have one channel. Use the non l_*, a_*, and b_* method.

RGB565 thresholds have three channels. Use the l_*, a_*, and b_* methods.

image.threshold

class image.threshold

Please call histogram.get_threshold() to create this object.

value

value() -> int

Return the grayscale threshold value (between 0 and 255).

You may also get this value doing [0] on the object.

l_value

l_value() -> int

Return the RGB565 LAB L channel threshold value (between 0 and 100).

You may also get this value doing [0] on the object.

a_value

a_value() -> int

Return the RGB565 LAB A channel threshold value (between -128 and 127).

You may also get this value doing [1] on the object.

b_value

b_value() -> int

Return the RGB565 LAB B channel threshold value (between -128 and 127).

You may also get this value doing [2] on the object.

class Statistics – Statistics Object

The percentile object is returned by histogram.get_statistics() or Image.get_statistics().

Grayscale statistics have one channel. Use the non l_*, a_*, and b_* method.

RGB565 statistics have three channels. Use the l_*, a_*, and b_* methods.

image.statistics

class image.statistics

Please call histogram.get_statistics() or Image.get_statistics() to create this object.

mean

mean() -> int

Returns the grayscale mean (0-255) (int).

You may also get this value doing [0] on the object.

median

median() -> int

Returns the grayscale median (0-255) (int).

You may also get this value doing [1] on the object.

mode

mode() -> int

Returns the grayscale mode (0-255) (int).

You may also get this value doing [2] on the object.

stdev

stdev() -> int

Returns the grayscale standard deviation (0-255) (int).

You may also get this value doing [3] on the object.

min

min() -> int

Returns the grayscale min (0-255) (int).

You may also get this value doing [4] on the object.

max

max() -> int

Returns the grayscale max (0-255) (int).

You may also get this value doing [5] on the object.

lq

lq() -> int

Returns the grayscale lower quartile (0-255) (int).

You may also get this value doing [6] on the object.

uq

uq() -> int

Returns the grayscale upper quartile (0-255) (int).

You may also get this value doing [7] on the object.

l_mean

l_mean() -> int

Returns the RGB565 LAB L mean (0-255) (int).

You may also get this value doing [0] on the object.

l_median

l_median() -> int

Returns the RGB565 LAB L median (0-255) (int).

You may also get this value doing [1] on the object.

l_mode

l_mode() -> int

Returns the RGB565 LAB L mode (0-255) (int).

You may also get this value doing [2] on the object.

l_stdev

l_stdev() -> int

Returns the RGB565 LAB L standard deviation (0-255) (int).

You may also get this value doing [3] on the object.

l_min

l_min() -> int

Returns the RGB565 LAB L min (0-255) (int).

You may also get this value doing [4] on the object.

l_max

l_max() -> int

Returns the RGB565 LAB L max (0-255) (int).

You may also get this value doing [5] on the object.

l_lq

l_lq() -> int

Returns the RGB565 LAB L lower quartile (0-255) (int).

You may also get this value doing [6] on the object.

l_uq

l_uq() -> int

Returns the RGB565 LAB L upper quartile (0-255) (int).

You may also get this value doing [7] on the object.

a_mean

a_mean() -> int

Returns the RGB565 LAB A mean (0-255) (int).

You may also get this value doing [8] on the object.

a_median

a_median() -> int

Returns the RGB565 LAB A median (0-255) (int).

You may also get this value doing [9] on the object.

a_mode

a_mode() -> int

Returns the RGB565 LAB A mode (0-255) (int).

You may also get this value doing [10] on the object.

a_stdev

a_stdev() -> int

Returns the RGB565 LAB A standard deviation (0-255) (int).

You may also get this value doing [11] on the object.

a_min

a_min() -> int

Returns the RGB565 LAB A min (0-255) (int).

You may also get this value doing [12] on the object.

a_max

a_max() -> int

Returns the RGB565 LAB A max (0-255) (int).

You may also get this value doing [13] on the object.

a_lq

a_lq() -> int

Returns the RGB565 LAB A lower quartile (0-255) (int).

You may also get this value doing [14] on the object.

a_uq

a_uq() -> int

Returns the RGB565 LAB A upper quartile (0-255) (int).

You may also get this value doing [15] on the object.

b_mean

b_mean() -> int

Returns the RGB565 LAB B mean (0-255) (int).

You may also get this value doing [16] on the object.

b_median

b_median() -> int

Returns the RGB565 LAB B median (0-255) (int).

You may also get this value doing [17] on the object.

b_mode

b_mode() -> int

Returns the RGB565 LAB B mode (0-255) (int).

You may also get this value doing [18] on the object.

b_stdev

b_stdev() -> int

Returns the RGB565 LAB B standard deviation (0-255) (int).

You may also get this value doing [19] on the object.

b_min

b_min() -> int

Returns the RGB565 LAB B min (0-255) (int).

You may also get this value doing [20] on the object.

b_max

b_max() -> int

Returns the RGB565 LAB B max (0-255) (int).

You may also get this value doing [21] on the object.

b_lq

b_lq() -> int

Returns the RGB565 LAB B lower quartile (0-255) (int).

You may also get this value doing [22] on the object.

b_uq

b_uq() -> int

Returns the RGB565 LAB B upper quartile (0-255) (int).

You may also get this value doing [23] on the object.

class Blob – Blob object

The blob object is returned by Image.find_blobs().

image.blob

class image.blob

Please call Image.find_blobs() to create this object.

corners

corners() -> List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners of the object. Corners are always returned in sorted clock-wise order starting from the top left.

min_corners

min_corners() -> List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners than bound the min area rectangle of the blob. Unlike blob.corners() the min area rectangle corners do not necessarily lie on the blob.

rect

rect() -> Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the blob’s bounding box.

x

x() -> int

Returns the blob’s bounding box x coordinate (int).

You may also get this value doing [0] on the object.

y

y() -> int

Returns the blob’s bounding box y coordinate (int).

You may also get this value doing [1] on the object.

w

w() -> int

Returns the blob’s bounding box w coordinate (int).

You may also get this value doing [2] on the object.

h

h() -> int

Returns the blob’s bounding box h coordinate (int).

You may also get this value doing [3] on the object.

pixels

pixels() -> int

Returns the number of pixels that are part of this blob (int).

You may also get this value doing [4] on the object.

cx

cx() -> int

Returns the centroid x position of the blob (int).

You may also get this value doing [5] on the object.

cxf

cxf() -> int

Returns the centroid x position of the blob (float).

cy

cy() -> int

Returns the centroid y position of the blob (int).

You may also get this value doing [6] on the object.

cyf

cyf() -> int

Returns the centroid y position of the blob (float).

rotation

rotation() -> float

Returns the rotation of the blob in radians (float). If the blob is like a pencil or pen this value will be unique for 0-180 degrees. If the blob is round this value is not useful.

You may also get this value doing [7] on the object.

rotation_deg

rotation_deg() -> float

Returns the rotation of the blob in degrees.

rotation_rad

rotation_rad() -> float

Returns the rotation of the blob in radians. This method is more descriptive than just blob.rotation().

code

code() -> int

Returns a 32-bit binary number with a bit set in it for each color threshold that’s part of this blob. For example, if you passed Image.find_blobs() three color thresholds to look for then bits 0/1/2 may be set for this blob. Note that only one bit will be set for each blob unless Image.find_blobs() was called with merge=True. Then its possible for multiple blobs with different color thresholds to be merged together. You can use this method along with multiple thresholds to implement color code tracking.

You may also get this value doing [8] on the object.

count

count() -> int

Returns the number of blobs merged into this blob. This is 1 unless you called Image.find_blobs() with merge=True.

You may also get this value doing [9] on the object.

perimeter

perimeter() -> int

Returns the number of pixels on this blob’s perimeter.

roundness

roundness() -> float

Returns a value between 0 and 1 representing how round the object is. A circle would be a 1.

elongation

elongation() -> float

Returns a value between 0 and 1 representing how long (not round) the object is. A line would be a 1.

area

area() -> int

Returns the area of the bounding box around the blob. (w * h).

density

density() -> float

Returns the density ratio of the blob. This is the number of pixels in the blob over its bounding box area. A low density ratio means in general that the lock on the object isn’t very good. The result is between 0 and 1.

extent

extent() -> float

Alias for blob.density().

compactness

compactness() -> float

Like blob.density(), but, uses the perimeter of the blob instead to measure the objects density and is thus more accurate. The result is between 0 and 1.

solidity

solidity() -> float

Like blob.density() but, uses the minimum area rotated rectangle versus the bounding rectangle to measure density. The result is between 0 and 1.

convexity

convexity() -> float

Returns a value between 0 and 1 representing how convex the object is. A square would be 1.

x_hist_bins

x_hist_bins() -> List[float]

Returns a histogram of the x axis of all columns in a blob. Bin values are scaled between 0 and 1.

y_hist_bins

y_hist_bins() -> List[float]

Returns a histogram of the y axis of all the rows in a blob. Bin values are scaled between 0 and 1.

major_axis_line

major_axis_line() -> Tuple[int, int, int, int]

Returns a line tuple (x1, y1, x2, y2) that can be drawn with Image.draw_line() of the major axis of the blob (the line going through the longest side of the min area rectangle).

minor_axis_line

minor_axis_line() -> Tuple[int, int, int, int]

Returns a line tuple (x1, y1, x2, y2) that can be drawn with Image.draw_line() of the minor axis of the blob (the line going through the shortest side of the min area rectangle).

enclosing_circle

enclosing_circle() -> Tuple[int, int, int]

Returns a circle tuple (x, y, r) that can be drawn with Image.draw_circle() of the circle that encloses the min area rectangle of a blob.

enclosed_ellipse

enclosed_ellipse() -> Tuple[int, int, int, int, float]

Returns an ellipse tuple (x, y, rx, ry, rotation) that can be drawn with Image.draw_ellipse() of the ellipse that fits inside of the min area rectangle of a blob.

class Line – Line object

The line object is returned by Image.find_lines(), Image.find_line_segments(), or Image.get_regression().

image.line

class image.line

Please call Image.find_lines(), Image.find_line_segments(), or Image.get_regression() to create this object.

line

line() -> Tuple[int, int, int, int]

Returns a line tuple (x1, y1, x2, y2) for use with other image methods like Image.draw_line().

x1

x1() -> int

Returns the line’s p1 x component.

You may also get this value doing [0] on the object.

y1

y1() -> int

Returns the line’s p1 y component.

You may also get this value doing [1] on the object.

x2

x2() -> int

Returns the line’s p2 x component.

You may also get this value doing [2] on the object.

y2

y2() -> int

Returns the line’s p2 y component.

You may also get this value doing [3] on the object.

length

length() -> int

Returns the line’s length: sqrt(((x2-x1)^2) + ((y2-y1)^2).

You may also get this value doing [4] on the object.

magnitude

magnitude() -> int

Returns the magnitude of the line from the hough transform.

You may also get this value doing [5] on the object.

theta

theta() -> int

Returns the angle of the line from the hough transform - (0 - 179) degrees.

You may also get this value doing [7] on the object.

rho

rho() -> int

Returns the the rho value for the line from the hough transform.

You may also get this value doing [8] on the object.

class Circle – Circle object

The circle object is returned by Image.find_circles().

image.circle

class image.circle

Please call Image.find_circles() to create this object.

x

x() -> int

Returns the circle’s x position.

You may also get this value doing [0] on the object.

y

y() -> int

Returns the circle’s y position.

You may also get this value doing [1] on the object.

r

r() -> int

Returns the circle’s radius.

You may also get this value doing [2] on the object.

magnitude

magnitude() -> int

Returns the circle’s magnitude.

You may also get this value doing [3] on the object.

class Rect – Rectangle Object

The rect object is returned by Image.find_rects().

image.rect

class image.rect

Please call Image.find_rects() to create this object.

corners

corners() -> List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners of the object. Corners are always returned in sorted clock-wise order starting from the top left.

rect

rect() -> Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the rect’s bounding box.

x

x() -> int

Returns the rectangle’s top left corner’s x position.

You may also get this value doing [0] on the object.

y

y() -> int

Returns the rectangle’s top left corner’s y position.

You may also get this value doing [1] on the object.

w

w() -> int

Returns the rectangle’s width.

You may also get this value doing [2] on the object.

h

h() -> int

Returns the rectangle’s height.

You may also get this value doing [3] on the object.

magnitude

magnitude() -> int

Returns the rectangle’s magnitude.

You may also get this value doing [4] on the object.

class QRCode – QRCode object

The qrcode object is returned by Image.find_qrcodes().

image.qrcode

class image.qrcode

Please call Image.find_qrcodes() to create this object.

corners

corners() -> List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners of the object. Corners are always returned in sorted clock-wise order starting from the top left.

rect

rect() -> Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the qrcode’s bounding box.

x

x() -> int

Returns the qrcode’s bounding box x coordinate (int).

You may also get this value doing [0] on the object.

y

y() -> int

Returns the qrcode’s bounding box y coordinate (int).

You may also get this value doing [1] on the object.

w

w() -> int

Returns the qrcode’s bounding box w coordinate (int).

You may also get this value doing [2] on the object.

h

h() -> int

Returns the qrcode’s bounding box h coordinate (int).

You may also get this value doing [3] on the object.

payload

payload() -> str

Returns the payload string of the qrcode. E.g. the URL.

You may also get this value doing [4] on the object.

version

version() -> int

Returns the version number of the qrcode (int).

You may also get this value doing [5] on the object.

ecc_level

ecc_level() -> int

Returns the ecc_level of the qrcode (int).

You may also get this value doing [6] on the object.

mask

mask() -> int

Returns the mask of the qrcode (int).

You may also get this value doing [7] on the object.

data_type

data_type() -> int

Returns the data type of the qrcode (int).

You may also get this value doing [8] on the object.

eci

eci() -> int

Returns the eci of the qrcode (int). The eci stores the encoding of data bytes in the QR Code. If you plan to handling QR Codes that contain more than just standard ASCII text you will need to look at this value.

You may also get this value doing [9] on the object.

is_numeric

is_numeric() -> bool

Returns True if the data_type of the qrcode is numeric.

is_alphanumeric

is_alphanumeric() -> bool

Returns True if the data_type of the qrcode is alpha numeric.

is_binary

is_binary() -> bool

Returns True if the data_type of the qrcode is binary. If you are serious about handling all types of text you need to check the eci if this is True to determine the text encoding of the data. Usually, it’s just standard ASCII, but, it could be UTF8 that has some 2-byte characters in it.

is_kanji

is_kanji() -> bool

Returns True if the data_type of the qrcode is alpha Kanji. If this is True then you’ll need to decode the string yourself as Kanji symbols are 10-bits per character and MicroPython has no support to parse this kind of text. The payload in this case must be treated as just a large byte array.

class AprilTag – AprilTag object

The apriltag object is returned by Image.find_apriltags().

image.apriltag

class image.apriltag

Please call Image.find_apriltags() to create this object.

Attributes

corners

corners: List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners of the object. Corners are always returned in sorted clock-wise order starting from the top left.

rect

rect: Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the apriltag’s bounding box.

x

x: int

Returns the apriltag’s bounding box x coordinate (int).

y

y: int

Returns the apriltag’s bounding box y coordinate (int).

w

w: int

Returns the apriltag’s bounding box w coordinate (int).

h

h: int

Returns the apriltag’s bounding box h coordinate (int).

area

area: int

Returns the area (w * h) of the apriltag (int).

id

id: int

Returns the numeric id of the apriltag.

TAG16H5 -> 0 to 29
TAG25H7 -> 0 to 241
TAG25H9 -> 0 to 34
TAG36H10 -> 0 to 2319
TAG36H11 -> 0 to 586
ARTOOLKIT -> 0 to 511

family

family: int

Returns the numeric family of the apriltag.

image.TAG16H5
image.TAG25H7
image.TAG25H9
image.TAG36H10
image.TAG36H11
image.ARTOOLKIT

name

name: str

Returns the family of the apriltag.

“TAG16H5”
“TAG25H7”
“TAG25H9”
“TAG36H10”
“TAG36H11”
“ARTOOLKIT”

cx

cx: int

Returns the centroid x position of the apriltag (int).

cxf

cxf: float

Returns the centroid x position of the apriltag (float).

cy

cy: int

Returns the centroid y position of the apriltag (int).

cyf

cyf: float

Returns the centroid y position of the apriltag (float).

rotation

rotation: float

Returns the rotation of the apriltag in radians (float).

decision_margin

decision_margin: float

Returns the quality of the apriltag match (0.0 - 1.0) where 1.0 is the best.

hamming

hamming: int

Returns the number of accepted bit errors for this tag.

TAG16H5 -> 0 bit errors will be accepted
TAG25H7 -> up to 1 bit error may be accepted
TAG25H9 -> up to 3 bit errors may be accepted
TAG36H10 -> up to 3 bit errors may be accepted
TAG36H11 -> up to 4 bit errors may be accepted
ARTOOLKIT -> 0 bit errors will be accepted

goodness

goodness: float

Returns the quality of the apriltag image (0.0 - 1.0) where 1.0 is the best.

Note

This value is always 0.0 for now. We may enable a feature called “tag refinement” in the future which will allow detection of small apriltags. However, this feature currently drops the frame rate to less than 1 FPS.

x_translation

x_translation: float

Returns the translation in unknown units from the camera in the X direction.

This method is useful for determining the apriltag’s location away from the camera. However, the size of the apriltag, the lens you are using, etc. all come into play as to actually determining what the X units are in. For ease of use we recommend you use a lookup table to convert the output of this method to something useful for your application.

Note that this is the left-to-right direction.

y_translation

y_translation: float

Returns the translation in unknown units from the camera in the Y direction.

This method is useful for determining the apriltag’s location away from the camera. However, the size of the apriltag, the lens you are using, etc. all come into play as to actually determining what the Y units are in. For ease of use we recommend you use a lookup table to convert the output of this method to something useful for your application.

Note that this is the up-to-down direction.

z_translation

z_translation: float

Returns the translation in unknown units from the camera in the Z direction.

This method is useful for determining the apriltag’s location away from the camera. However, the size of the apriltag, the lens you are using, etc. all come into play as to actually determining what the Z units are in. For ease of use we recommend you use a lookup table to convert the output of this method to something useful for your application.

Note that this is the front-to-back direction.

x_rotation

x_rotation: float

Returns the rotation in radians of the apriltag in the X plane. E.g. moving the camera left-to-right while looking at the tag.

y_rotation

y_rotation: float

Returns the rotation in radians of the apriltag in the Y plane. E.g. moving the camera up-to-down while looking at the tag.

z_rotation

z_rotation: float

Returns the rotation in radians of the apriltag in the Z plane. E.g. rotating the camera while looking directly at the tag.

Note that this is just a renamed version of apriltag.rotation.

class DataMatrix – DataMatrix object

The datamatrix object is returned by Image.find_datamatrices().

image.datamatrix

class image.datamatrix

Please call Image.find_datamatrices() to create this object.

corners

corners() -> List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners of the object. Corners are always returned in sorted clock-wise order starting from the top left.

rect

rect() -> Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the datamatrix’s bounding box.

x

x() -> int

Returns the datamatrix’s bounding box x coordinate (int).

You may also get this value doing [0] on the object.

y

y() -> int

Returns the datamatrix’s bounding box y coordinate (int).

You may also get this value doing [1] on the object.

w

w() -> int

Returns the datamatrix’s bounding box w coordinate (int).

You may also get this value doing [2] on the object.

h

h() -> int

Returns the datamatrix’s bounding box h coordinate (int).

You may also get this value doing [3] on the object.

payload

payload() -> str

Returns the payload string of the datamatrix. E.g. The string.

You may also get this value doing [4] on the object.

rotation

rotation() -> float

Returns the rotation of the datamatrix in radians (float).

You may also get this value doing [5] on the object.

rows

rows() -> int

Returns the number of rows in the data matrix (int).

You may also get this value doing [6] on the object.

columns

columns() -> int

Returns the number of columns in the data matrix (int).

You may also get this value doing [7] on the object.

capacity

capacity() -> int

Returns how many characters could fit in this data matrix.

You may also get this value doing [8] on the object.

padding

padding() -> int

Returns how many unused characters are in this data matrix.

You may also get this value doing [9] on the object.

class BarCode – BarCode object

The barcode object is returned by Image.find_barcodes().

image.barcode

class image.barcode

Please call Image.find_barcodes() to create this object.

corners

corners() -> List[Tuple[int, int]]

Returns a list of 4 (x,y) tuples of the 4 corners of the object. Corners are always returned in sorted clock-wise order starting from the top left.

rect

rect() -> Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the barcode’s bounding box.

x

x() -> int

Returns the barcode’s bounding box x coordinate (int).

You may also get this value doing [0] on the object.

y

y() -> int

Returns the barcode’s bounding box y coordinate (int).

You may also get this value doing [1] on the object.

w

w() -> int

Returns the barcode’s bounding box w coordinate (int).

You may also get this value doing [2] on the object.

h

h() -> int

Returns the barcode’s bounding box h coordinate (int).

You may also get this value doing [3] on the object.

payload

payload() -> str

Returns the payload string of the barcode. E.g. The number.

You may also get this value doing [4] on the object.

type

type() -> int

Returns the type enumeration of the barcode (int).

You may also get this value doing [5] on the object.

image.EAN2
image.EAN5
image.EAN8
image.UPCE
image.ISBN10
image.UPCA
image.EAN13
image.ISBN13
image.I25
image.DATABAR
image.DATABAR_EXP
image.CODABAR
image.CODE39
image.PDF417 - Future (e.g. doesn’t work right now).
image.CODE93
image.CODE128

rotation

rotation() -> float

Returns the rotation of the barcode in radians (float).

You may also get this value doing [6] on the object.

quality

quality() -> int

Returns the number of times this barcode was detected in the image (int).

When scanning a barcode each new scanline can decode the same barcode. This value increments for a barcode each time that happens…

You may also get this value doing [7] on the object.

class Displacement – Displacement object

The displacement object is returned by Image.find_displacement().

image.displacement

class image.displacement

Please call Image.find_displacement() to create this object.

x_translation

x_translation() -> float

Returns the x translation in pixels between two images. This is sub pixel accurate so it’s a float.

You may also get this value doing [0] on the object.

y_translation

y_translation() -> float

Returns the y translation in pixels between two images. This is sub pixel accurate so it’s a float.

You may also get this value doing [1] on the object.

rotation

rotation() -> float

Returns the rotation in radians between two images.

You may also get this value doing [2] on the object.

scale

scale() -> float

Returns the scale change between two images.

You may also get this value doing [3] on the object.

response

response() -> float

Returns the quality of the results of displacement matching between two images. Between 0-1. A displacement object with a response less than 0.1 is likely noise.

You may also get this value doing [4] on the object.

class kptmatch – Keypoint Object

The kptmatch object is returned by image.match_descriptor() for keypoint matches.

image.kptmatch

class image.kptmatch

Please call image.match_descriptor() to create this object.

rect

rect() -> Tuple[int, int, int, int]

Returns a rectangle tuple (x, y, w, h) for use with other image methods like Image.draw_rectangle() of the kptmatch’s bounding box.

cx

cx() -> int

Returns the centroid x position of the kptmatch (int).

You may also get this value doing [0] on the object.

cy

cy() -> int

Returns the centroid y position of the kptmatch (int).

You may also get this value doing [1] on the object.

x

x() -> int

Returns the kptmatch’s bounding box x coordinate (int).

You may also get this value doing [2] on the object.

y

y() -> int

Returns the kptmatch’s bounding box y coordinate (int).

You may also get this value doing [3] on the object.

w

w() -> int

Returns the kptmatch’s bounding box w coordinate (int).

You may also get this value doing [4] on the object.

h

h() -> int

Returns the kptmatch’s bounding box h coordinate (int).

You may also get this value doing [5] on the object.

count

count() -> int

Returns the number of keypoints matched (int).

You may also get this value doing [6] on the object.

theta

theta() -> int

Returns the estimated angle of rotation for the keypoint (int).

You may also get this value doing [7] on the object.

match

match() -> List[Tuple[int, int]]

Returns the list of (x,y) tuples of matching keypoints.

You may also get this value doing [8] on the object.

class ImageIO – ImageIO Object

The ImageIO object allows you to read/write OpenMV Image objects in their native form to disk or to memory. This class provides fast read/write random access for loading/storing images.

image.ImageIO

class image.ImageIO(path: str, mode)

Creates an ImageIO object.

If path is a file name on disk then that file will be opened for reading if mode is 'r' or writing if mode is 'w'.

path may also be a 3-value tuple (w, h, bpp) for in-memory storage of images. mode in this case is then the number of image buffers to store in memory. Note that the in-memory storage buffer is not allowed to grow in size after being allocated. Use a bpp value of 0 for binary images, 1 for grayscale images, and 2 for rgb565 images.

type

type() -> int

Returns if the ImageIO object is a FILE_STREAM or MEMORY_STREAM.

is_closed

is_closed() -> bool

Returns if the ImageIO object is closed and can no longer be used.

count

count() -> int

Returns the number of frames stored.

offset

offset() -> int

Returns the image index offset.

version

version() -> int | None

Returns the version of the object if it’s FILE_STREAM. MEMORY_STREAM versions are none.

buffer_size

buffer_size() -> int

Returns the size allocated by the object for a frame in a single buffer.

buffer_size() * count() == size()

size

size() -> int

Returns the number of bytes on disk or memory used by the ImageIO object.

write

write(img: Image) -> ImageIO

Writes a new image img to the ImageIO object. For on disk ImageIO objects the file will grow as new images are added. For in-memory ImageIO objects this just writes an image to the current pre-allocated slot before advancing to the next slot.

Returns the ImageIO object.

read

read(copy_to_fb=True, loop=True, pause=True) -> Image

Returns an image object from the ImageIO object. If copy_to_fb is False then the new image is allocated on the MicroPython heap. However, the MicroPython heap is limited and may not have space to store the new image if exhausted. Instead, set copy_to_fb to True to set the frame buffer to the new image making this function work just like sensor.snapshot().

loop if True automatically causes the ImageIO object to seek to the beginning at the end of the stream of images.

pause if True causes this method to pause for a previously recorded number of milliseconds by write in-order to match the original frame rate that captured the image data.

seek

seek(offset) -> None

Seeks to the image slot number offset in the ImageIO object.

Works for on disk or in-memory objects.

sync

sync() -> None

Writes out all data pending for on-disk ImageIO objects.

close

close() -> None

Closes the ImageIO object. For in-memory objects this free’s the allocated space and for on-disk files this closes the file and writes out all meta-data.

FILE_STREAM

FILE_STREAM: int

ImageIO object was opened on a file.

MEMORY_STREAM

MEMORY_STREAM: int

ImageIO object was opened in memory.

class Image – Image object

The image object is the basic object for machine vision operations.

image.Image

class image.Image(arg, buffer: bytes | bytearray | memoryview | None = None, copy_to_fb: bool = False)

If arg is a string then this creates a new image object from a file at arg path. Supports loading bmp/pgm/ppm/jpg/jpeg/png image files from disk. If copy_to_fb is true the image is copied to the frame buffer verus being allocated on the heap.

If arg is an ndarray then this creates a new image object from the ndarray. ndarray objects with a shape of (w, h) are treated as grayscale images, (w, h, 3) are treated as RGB565 images. Only float32 point ndarrays are supported at this time. When creating an image this way if you pass a buffer argument it will be used to store the image data versus allocating space on the heap. If copy_to_fb is true the image is copied to the frame buffer verus being allocated on the heap or using the buffer.

If arg is an int it is then considered the width of a new image and a height value and a format value must follow to create a new blank image object. format can be be any image pixformat value like image.GRAYSCALE. The image will be initialized to all zeros. Note that a buffer value is expected for compressed image formats. buffer is considered as the source of image data for creating images this way. If used with copy_to_fb the data from buffer is copied to the frame buffer. If you’d like to create a JPEG image from a JPEG bytes() or bytearray() object you can pass the width, height, image.JPEG for the JPEG along with setting buffer to the JPEG byte stream to create a JPEG image.

Images support “[]” notation. Do image[index] = 8/16-bit value to assign an image pixel or image[index] to get an image pixel which will be either an 8-bit value for grayscale/bayer images of a 16-bit value for RGB565/YUV images. Binary images return a 1-bit value.

For JPEG images the “[]” allows you to access the compressed JPEG image blob as a byte-array. Reading and writing to the data array is opaque however as JPEG images are compressed byte streams.

Images also support read buffer operations. You can pass images to all sorts of MicroPython functions like as if the image were a byte-array object. In particular, if you’d like to transmit an image you can just pass it to the UART/SPI/I2C write functions to be transmitted automatically.

Basic Methods

width

width() -> int

Returns the image width in pixels.

height

height() -> int

Returns the image height in pixels.

format

format() -> int

Returns image.GRAYSCALE for grayscale images, image.RGB565 for RGB565 images, image.BAYER for bayer pattern images, and image.JPEG for JPEG images.

size

size() -> int

Returns the image size in bytes.

bytearray

bytearray() -> bytearray

Returns a bytearray object that points to the image data for byte-level read/write access.

Note

Image objects are automatically cast as bytes objects when passed to MicroPython driver that requires a bytes like object. This is read-only access. Call bytearray() to get read/write access.

get_pixel

get_pixel(x: int, y: int, rgbtuple: bool | None = None) -> int | Tuple[int, int, int]

For grayscale images: Returns the grayscale pixel value at location (x, y). For RGB565 images: Returns the RGB888 pixel tuple (r, g, b) at location (x, y). For bayer pattern images: Returns the the pixel value at the location (x, y).

Returns None if x or y is outside of the image.

x and y may either be passed independently or as a tuple.

rgbtuple if True causes this method to return an RGB888 tuple. Otherwise, this method returns the integer value of the underlying pixel. I.e. for RGB565 images this method returns a RGB565 value. Defaults to True for RGB565 images and False otherwise.

Not supported on compressed images.

Note

Image.get_pixel() and Image.set_pixel() are the only methods that allow you to manipulate bayer pattern images. Bayer pattern images are literal images where pixels in the image are R/G/R/G/etc. for even rows and G/B/G/B/etc. for odd rows. Each pixel is 8-bits. If you call this method with rgbtuple set then Image.get_pixel() will debayer the source image at that pixel location and return a valid RGB888 tuple for the pixel location.

set_pixel

set_pixel(x: int, y: int, pixel: int | Tuple[int, int, int]) -> Image

For grayscale images: Sets the pixel at location (x, y) to the grayscale value pixel. For RGB565 images: Sets the pixel at location (x, y) to the RGB888 tuple (r, g, b) pixel. For bayer pattern images: Sets the pixel value at the location (x, y) to the value pixel.

Returns the image object so you can call another method using . notation.

x and y may either be passed independently or as a tuple.

pixel may either be an RGB888 tuple (r, g, b) or the underlying pixel value (i.e. a RGB565 value for RGB565 images or an 8-bit value for grayscale images.

Not supported on compressed images.

Note

Image.get_pixel() and Image.set_pixel() are the only methods that allow you to manipulate bayer pattern images. Bayer pattern images are literal images where pixels in the image are R/G/R/G/etc. for even rows and G/B/G/B/etc. for odd rows. Each pixel is 8-bits. If you call this method with an RGB888 tuple the grayscale value of that RGB888 tuple is extracted and set to the pixel location.

Conversion Methods

to_ndarray

to_ndarray(dtype: str, buffer: bytes | bytearray | memoryview | None = None) -> ndarray

Returns a ndarray object created from the image. This only works for GRAYSCALE or RGB565 images currently.

dtype can be b, B, or f for creating a signed 8-bit, unsigned 8-bit, or 32-bit floating point ndarray. GRAYSCALE images are directly converted to unsigned 8-bit ndarray objects. For signed 8-bit ndarray objects the values (0:255) are mapped to (-127:128). For float 32-bit ndarray objects the values are mapped to (0.0:255.0). RGB565 images are converted to 3-channel ndarray objects and the same process described above for GRAYSCALE images is applied to each channel depending on dtype. Note that dtype also accepts the integer values (e.g. ord()) of b, B, and f respectively.

buffer if not None is a bytearray object to use as the buffer for the ndarray. If None a new buffer is allocated on the heap to store the ndarray image data. You can use the buffer argument to directly allocate the ndarray in a pre-allocated buffer saving a heap allocation and a copy operation.

The ndarray returned has the shape of (height, width) for GRAYSCALE images and (height, width, 3) for RGB565 images.

to_bitmap

to_bitmap(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Converts an image to a bitmap image (1 bit per pixel).

x_scale controls how much the image is scaled by in the x direction (float). If this value is negative the image will be flipped horizontally. Note that if y_scale is not specified then it will match x_scale to maintain the aspect ratio.

y_scale controls how much the image is scaled by in the y direction (float). If this value is negative the image will be flipped vertically. Note that if x_scale is not specified then it will match x_scale to maintain the aspect ratio.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

rgb_channel is the RGB channel (0=R, G=1, B=2) to extract from an RGB565 image (if passed) and to apply onto the final image. For example, if you pass rgb_channel=1 this will extract the green channel of the source RGB565 image and apply that in grayscale on the final image.

alpha controls how much of the source image to blend into the final image. A value of 256 draws an opaque source image while a value lower than 256 produces a blend between the original and final image. 0 results in no modification to the final image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the image is. This is applied after rgb_channel extraction if used.

alpha_palette if not None can be a 256 pixel in total GRAYSCALE image to use as a alpha palette which modulates the alpha value of the source image being drawn at a pixel pixel level allowing you to precisely control the alpha value of pixels based on their grayscale value. A pixel value of 255 in the alpha lookup table is opaque which anything less than 255 becomes more transparent until 0. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Note

Bitmap images are like grayscale images with only two pixels values - 0 and 1. Additionally, bitmap images are packed such that they only store 1 bit per pixel making them very small. The OpenMV image library allows bitmap images to be used in all places sensor.GRAYSCALE and sensor.RGB565 images can be used. However, many operations when applied on bitmap images don’t make any sense because bitmap images only have 2 values. OpenMV recommends using bitmap images for mask values in operations and such as they fit on the MicroPython heap quite easily. Finally, bitmap image pixel values 0 and 1 are interpreted as black and white when being applied to sensor.GRAYSCALE or sensor.RGB565 images. The library automatically handles conversion.

Returns the image object so you can call another method using . notation.

to_grayscale

to_grayscale(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, copy: bool = False, copy_to_fb: bool = False) -> Image

Converts an image to a grayscale image (8-bits per pixel).

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

to_rgb565

to_rgb565(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Converts an image to an RGB565 image (16-bits per pixel).

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

to_rainbow

to_rainbow(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=PALETTE_RAINBOW, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Converts an image to an RGB565 rainbow image (16-bits per pixel).

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

to_ironbow

to_ironbow(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=PALETTE_IRONBOW, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Converts an image to an RGB565 ironbow image (16-bits per pixel).

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

to_jpeg

to_jpeg(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False, quality: int = 90, encode_for_ide: bool = False, subsampling: int = 0) -> Image

Converts an image to a JPEG image.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

quality controls the jpeg image compression quality. The value can be between 0 and 100.

encode_for_ide if True the image is encoded in a way that the IDE can display it if printed by doing print(image). This is useful for debugging purposes over UARTs via Open Terminal in the IDE.

subsampling can be:

image.JPEG_SUBSAMPLING_AUTO: Use the best subsampling for the image based on the quality.
image.JPEG_SUBSAMPLING_444: Use 4:4:4 subsampling.
image.JPEG_SUBSAMPLING_422: Use 4:2:2 subsampling.
image.JPEG_SUBSAMPLING_420: Use 4:2:0 subsampling.

Returns the image object so you can call another method using . notation.

to_png

to_png(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Converts an image to a PNG image.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

compress

compress(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False, quality: int = 90, encode_for_ide: bool = False, subsampling: int = 0) -> Image

Converts an image to a JPEG image.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

quality controls the jpeg image compression quality. The value can be between 0 and 100.

encode_for_ide if True the image is encoded in a way that the IDE can display it if printed by doing print(image). This is useful for debugging purposes over UARTs via Open Terminal in the IDE.

subsampling can be:

image.JPEG_SUBSAMPLING_AUTO: Use the best subsampling for the image based on the quality.
image.JPEG_SUBSAMPLING_444: Use 4:4:4 subsampling.
image.JPEG_SUBSAMPLING_422: Use 4:2:2 subsampling.
image.JPEG_SUBSAMPLING_420: Use 4:2:0 subsampling.

Returns the image object so you can call another method using . notation.

Note

Image.compress is an alias for Image.to_jpeg.

copy

copy(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy_to_fb: float = False) -> Image

Creates a deep copy of the image object.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy_to_fb if True the image is loaded directly into the frame buffer. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

crop

crop(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Modifies an image in-place without changing the underlying image type.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

scale

scale(x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, copy: bool = False, copy_to_fb: bool = False) -> Image

Modifies an image in-place without changing the underlying image type.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the image to extract. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the image.
image.VFLIP: Vertically flip the image.
image.TRANSPOSE: Transpose the image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.ROTATE_90: Rotate the image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the image by 270 degrees (this is just HMIRROR | TRANSPOSE).

copy if True create a deep-copy on the heap of the image that’s been converted versus converting the original image in-place.

copy_to_fb if True the image is loaded directly into the frame buffer. copy_to_fb has priority over copy. This has no special effect if the image is already in the frame buffer.

Returns the image object so you can call another method using . notation.

Note

Image.scale is an alias for Image.crop.

save

save(path: str, roi: Tuple[int, int, int, int] | None = None, quality=50) -> Image

Saves a copy of the image to the filesystem at path.

Supports bmp/pgm/ppm/jpg/jpeg image files. Note that you cannot save jpeg compressed images to an uncompressed format.

roi is the region-of-interest rectangle (x, y, w, h) to save from. If not specified, it is equal to the image rectangle which copies the entire image. This argument is not applicable for JPEG images.

quality is the jpeg compression quality to use to save the image to jpeg format if the image is not already compressed (0-100) (int).

Returns the image object so you can call another method using . notation.

flush

flush() -> None

Updates the frame buffer in the IDE with the image in the frame buffer on the camera.

Drawing Methods

clear

clear(mask: Image | None = None) -> Image

Sets all pixels in the image to zero (very fast).

mask is another image to use as a pixel level mask for the operation. The mask should be an image with just black or white pixels and should be the same size as the image being operated on. Only pixels set in the mask are modified.

Returns the image object so you can call another method using . notation.

Not supported on compressed images.

draw_line

draw_line(x0: int, y0: int, x1: int, y1: int, color: int | Tuple[int, int, int] | None = None, thickness=1) -> Image

Draws a line from (x0, y0) to (x1, y1) on the image. You may either pass x0, y0, x1, y1 separately or as a tuple (x0, y0, x1, y1).

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

thickness controls how thick the line is in pixels.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_rectangle

draw_rectangle(x: int, y: int, w: int, h: int, color: int | Tuple[int, int, int] | None = None, thickness=1, fill=False) -> Image

Draws a rectangle on the image. You may either pass x, y, w, h separately or as a tuple (x, y, w, h).

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

thickness controls how thick the lines are in pixels.

Pass fill set to True to fill the rectangle.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_circle

draw_circle(x: int, y: int, radius: int, color: int | Tuple[int, int, int] | None = None, thickness=1, fill=False) -> Image

Draws a circle on the image. You may either pass x, y, radius separately or as a tuple (x, y, radius).

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

thickness controls how thick the edges are in pixels.

Pass fill set to True to fill the circle.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_ellipse

draw_ellipse(cx: int, cy: int, rx: int, ry: int, rotation: int, color: int | Tuple[int, int, int] | None = None, thickness=1, fill=False) -> Image

Draws an ellipse on the image. You may either pass cx, cy, rx, ry, and the rotation (in degrees) separately or as a tuple (cx, yc, rx, ry, rotation).

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

thickness controls how thick the edges are in pixels.

Pass fill set to True to fill the ellipse.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_string

draw_string(x: int, y: int, text: str, color: int | Tuple[int, int, int] | None = None, scale=1, x_spacing=0, y_spacing=0, mono_space=True, char_rotation=0, char_hmirror=False, char_vflip=False, string_rotation=0, string_hmirror=False, string_vflip=False) -> Image

Draws 8x10 text starting at location (x, y) in the image. You may either pass x, y separately or as a tuple (x, y).

text is a string to write to the image. \n, \r, and \r\n line endings move the cursor to the next line.

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

scale may be increased to increase/decrease the size of the text on the image. You can pass greater than 0 integer or floating point values.

x_spacing allows you to add (if positive) or subtract (if negative) x pixels between characters.

y_spacing allows you to add (if positive) or subtract (if negative) y pixels between characters (for multi-line text).

mono_space defaults to True which forces text to be fixed spaced. For large text scales this looks terrible. Set the False to get non-fixed width character spacing which looks A LOT better.

char_rotation may be 0, 90, 180, 270 to rotate each character in the string by this amount.

char_hmirror if True horizontally mirrors all characters in the string.

char_vflip if True vertically flips all characters in the string.

string_rotation may be 0, 90, 180, 270 to rotate the string by this amount.

string_hmirror if True horizontally mirrors the string.

string_vflip if True vertically flips the string.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_cross

draw_cross(x: int, y: int, color: int | Tuple[int, int, int] | None = None, size=5, thickness=1) -> Image

Draws a cross at location x, y. You may either pass x, y separately or as a tuple (x, y).

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

size controls how long the lines of the cross extend.

thickness controls how thick the edges are in pixels.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_arrow

draw_arrow(x0: int, y0: int, x1: int, y1: int, color: int | Tuple[int, int, int] | None = None, thickness=1) -> Image

Draws an arrow from (x0, y0) to (x1, y1) on the image. You may either pass x0, y0, x1, y1 separately or as a tuple (x0, y0, x1, y1).

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

thickness controls how thick the line is in pixels.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_edges

draw_edges(image: Image, corners, color: int | Tuple[int, int, int] | None = None, size=0, thickness=1, fill=False) -> Image

Draws line edges between a corner list returned by methods like blob.corners. Corners is a four valued tuple of two valued x/y tuples. E.g. [(x1,y1),(x2,y2),(x3,y3),(x4,y4)].

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

size if greater than 0 causes the corners to be drawn as circles of radius size.

thickness controls how thick the line is in pixels.

Pass fill set to True to fill the corner circles if drawn.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

draw_image

draw_image(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Draws an image whose top-left corner starts at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. draw_image("test.jpg").

x_scale controls how much the source image is scaled by in the x direction (float). If this value is negative the image will be flipped horizontally. Note that if y_scale is not specified then it will match x_scale to maintain the aspect ratio.

y_scale controls how much the source image is scaled by in the y direction (float). If this value is negative the image will be flipped vertically. Note that if x_scale is not specified then it will match x_scale to maintain the aspect ratio.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

rgb_channel is the RGB channel (0=R, G=1, B=2) to extract from an RGB565 image (if passed) and to apply onto the destination image. For example, if you pass rgb_channel=1 this will extract the green channel of the source RGB565 image and apply that in grayscale on the destination image.

alpha controls how much of the source image to blend into the destination image. A value of 256 draws an opaque source image while a value lower than 256 produces a blend between the source and destination image. 0 results in no modification to the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

draw_keypoints

draw_keypoints(keypoints, color: int | Tuple[int, int, int] | None = None, size=10, thickness=1, fill=False) -> Image

Draws the keypoints of a keypoints object on the image. You may also pass a list of three value tuples containing the (x, y, rotation_angle_in_degrees) to reuse this method for drawing keypoint glyphs which are a circle with a line pointing in a particular direction.

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

size controls how large the keypoints are.

thickness controls how thick the line is in pixels.

Pass fill set to True to fill the keypoints.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

flood_fill

flood_fill(x: int, y: int, seed_threshold=0.05, floating_threshold=0.05, color: int | Tuple[int, int, int] | None = None, invert=False, clear_background=False, mask: Image | None = None) -> Image

Flood fills a region of the image starting from location x, y. You may either pass x, y separately or as a tuple (x, y).

seed_threshold controls how different any pixel in the fill area may be from the original starting pixel.

floating_threshold controls how different any pixel in the fill area may be from any neighbor pixels.

color is an RGB888 tuple for Grayscale or RGB565 images. Defaults to white. However, you may also pass the underlying pixel value (0-255) for grayscale images or a RGB565 value for RGB565 images.

Pass invert as True to re-color everything outside of the flood-fill connected area.

Pass clear_background as True to zero the rest of the pixels that flood-fill did not re-color.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

This method is not available on the OpenMV Cam M4.

Masking Methods

mask_rectange

mask_rectange(x: int, y: int, w: int, h: int) -> Image

Zeros a rectangular part of the image. If no arguments are supplied this method zeros the center of the image.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

mask_circle

mask_circle(x: int, y: int, radius: int) -> Image

Zeros a circular part of the image. If no arguments are supplied this method zeros the center of the image.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

mask_ellipse

mask_ellipse(x: int, y: int, radius_x: int, radius_y: int, rotation_angle_in_degrees: int) -> Image

Zeros an ellipsed shaped part of the image. If no arguments are supplied this method zeros the center of the image.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

Binary Methods

binary

binary(thresholds: List[Tuple[int, int]], invert=False, zero=False, mask: Image | None = None, to_bitmap=False, copy=False) -> Image

Sets all pixels in the image to black or white depending on if the pixel is inside of a threshold in the threshold list thresholds or not.

which are minimums and maximums for the LAB L, A, and B channels respectively. For easy usage this function will automatically fix swapped min and max values. Additionally, if a tuple is larger than six values the rest are ignored. Conversely, if the tuple is too short the rest of the thresholds are assumed to be at maximum range.

Note

To get the thresholds for the object you want to track just select (click and drag) on the object you want to track in the IDE frame buffer. The histogram will then update to just be in that area. Then just write down where the color distribution starts and falls off in each histogram channel. These will be your low and high values for thresholds. It’s best to manually determine the thresholds versus using the upper and lower quartile statistics because they are too tight.

You may also determine color thresholds by going into Tools->Machine Vision->Threshold Editor in OpenMV IDE and selecting thresholds from the GUI slider window.

invert inverts the thresholding operation such that instead of matching pixels inside of some known color bounds pixels are matched that are outside of the known color bounds.

Set zero to True to instead zero thresholded pixels and leave pixels not in the threshold list untouched.

to_bitmap turns the image data into a binary bitmap image where each pixel is stored in 1 bit. For very small images the new bitmap image may not fit inside of the original image requiring an out-of-place operation using copy.

copy if True creates a copy of the binarized image on the heap versus modifying the source image.

Note

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

invert

invert() -> Image

Flips (binary inverts) all pixels values in the image. Note that binary inversion is the same as numerical inversion for images because:

(255 - pixel) % 256 == (255 + ~pixel + 1) % 256 == (~pixel + 256) % 256 == ~pixel and this holds for any value that’s in a range of (0-2^n-1) which is true for all mutable image datatypes.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

b_and

b_and(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Finds the logical AND of image and this image (e.g. a & b), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. b_and("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

b_nand

b_nand(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Finds the logical NAND of image and this image (e.g. ~(a & b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. b_nand("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

b_or

b_or(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Finds the logical OR of image and this image (e.g. (a | b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. b_or("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

b_nor

b_nor(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Finds the logical NOR of image and this image (e.g. ~(a | b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. b_nor("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

b_xor

b_xor(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Finds the logical XOR of image and this image (e.g. (a ^ b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. b_xor("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

b_xnor

b_xnor(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Finds the logical XNOR of image and this image (e.g. ~(a ^ b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. b_xnor("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

ISP Methods

awb

awb(max: bool = False) -> Image

Performs automatic white balance on the image using the gray-world algorithm. This method operates on RAW Bayer Images so that you can improve image quality before converting to RGB565 or passing the RAW Bayer Image to an image processing function. You may also call this on a RGB565. This method has no affect on binary/grayscale images.

max if True uses the white-patch algorithm instead.

Returns the image object so you can call another method using . notation.

Not supported on compressed or yuv images.

ccm

ccm(matrix) -> Image

Multiples the passed floating-point color-correction-matrix with the image. Matrices may be in the form of:

[[rr, rg, rb], [gr, gg, gb], [br, bg, bb]]
[[rr, rg, rb], [gr, gg, gb], [br, bg, bb], [xx, xx, xx]]
[[rr, rg, rb, ro], [gr, gg, gb, go], [br, bg, bb, bo]]
[[rr, rg, rb, ro], [gr, gg, gb, go], [br, bg, bb, bo], [xx, xx, xx, xx]]

[rr, rg, rb, ro, gr, gg, gb, go, br, bg, bb, bo]
[rr, rg, rb, ro, gr, gg, gb, go, br, bg, bb, bo, xx, xx, xx, xx]

The CCM Method does:

|R'|                |R|      |R'|                |R|
|G'| = 3x3 Matrix * |G|  or  |G'| = 3x4 Matrix * |G|
|B'|                |B|      |B'|                |B|
                                                 |1|

Note that the sum of each row in the 3x3 matrix should generally be -1, +1, or 0. Weights may either be positive or negative.

You may want to use this method to eliminate systemic cross talk between color channels. Or alternatively, to do color correction on the whole image.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

gamma

gamma(gamma: float = 1.0, contrast: float = 1.0, brightness: float = 0.0) -> Image

Quickly changes the image gamma, contrast, and brightness.

gamma with values greater than 1.0 makes the image darker in a non-linear manner while less than 1.0 makes the image brighter. The gamma value is applied to the image by scaling all pixel color channels to be between [0:1) and then doing a remapping of pow(pixel, 1/gamma) on all pixels before scaling back.

contrast with values greater than 1.0 makes the image brighter in a linear manner while less than 1.0 makes the image darker. The contrast value is applied to the image by scaling all pixel color channels to be between [0:1) and then doing a remapping of pixel * contrast on all pixels before scaling back.

brightness with values greater than 0.0 makes the image brighter in a constant manner while less than 0.0 makes the image darker. The brightness value is applied to the image by scaling all pixel color channels to be between [0:1) and then doing a remapping of pixel + brightness on all pixels before scaling back.

Returns the image object so you can call another method using . notation.

Not supported on compressed or bayer/yuv images.

gamma_corr

gamma_corr(gamma: float = 1.0, contrast: float = 1.0, brightness: float = 0.0) -> Image

Quickly changes the image gamma, contrast, and brightness.

Returns the image object so you can call another method using . notation.

Not supported on compressed or bayer/yuv images.

Note

Image.gamma_corr is an alias for Image.gamma.

Math Methods

negate

negate() -> Image

Flips (binary inverts) all pixels values in the image. Note that binary inversion is the same as numerical inversion for images because:

(255 - pixel) % 256 == (255 + ~pixel + 1) % 256 == (~pixel + 256) % 256 == ~pixel and this holds for any value that’s in a range of (0-2^n-1) which is true for all mutable image datatypes.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

Note

Image.negate is an alias for Image.invert.

replace

replace(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

Note

Image.replace is an alias for Image.draw_image.

assign

assign(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

Note

Image.assign is an alias for Image.draw_image.

set

set(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

Note

Image.set is an alias for Image.draw_image.

add

add(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Numerically adds image and this image (e.g. min(a + b, 255)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. add("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

sub

sub(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Numerically subtracts image and this image (e.g. max(a - b, 0)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. sub("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

rsub

rsub(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Numerically reverse subtracts image and this image (e.g. max(b - a, 0)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. rsub("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

min

min(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Numerically finds the minimum of image and this image (e.g. min(a, b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. min("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

max

max(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Numerically finds the maximum of image and this image (e.g. max(a, b)), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. max("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

difference

difference(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

Numerically finds the absolute difference of image and this image (e.g. |a - b|), color channel by color channel, from the top-left corner at location x, y. This method automatically handles rendering the image passed into the correct pixel format for the destination image while also handling clipping seamlessly. image may also be a RGB888 tuple to draw a color instead of an image. You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. difference("test.jpg").

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

blend

blend(image: Image, x: int = 0, y: int = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, mask: Image | None = None) -> Image

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image to draw. This allows you to extract just the pixels in the ROI to scale and draw on the destination image.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being drawn on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being drawn to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being drawn to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being drawn to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being drawn on is black speeding up blending.

mask is another image to use as a pixel level mask for the operation.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

Note

Image.blend is an alias for Image.draw_image.

histeq

histeq(adaptive=False, clip_limit=-1, mask: Image | None = None) -> Image

Runs the histogram equalization algorithm on the image. Histogram equalization normalizes the contrast and brightness in the image.

If you pass adaptive as True then an adaptive histogram equalization method will be run on the image instead which as generally better results than non-adaptive histogram qualization but a longer run time.

clip_limit provides a way to limit the contrast of the adaptive histogram qualization. Use a small value for this, like 10, to produce good histogram equalized contrast limited images.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

Filtering Methods

erode

erode(size: int, threshold: int | None = None, mask: Image | None = None) -> Image

Removes pixels from the edges of segmented areas.

This method works by convolving a kernel of ((size*2)+1)x((size*2)+1) pixels across the image and zeroing the center pixel of the kernel if the sum of the neighbour pixels clear is greater than threshold.

This method works like the standard erode method if threshold is not set. If threshold is set then you can specify erode to only erode pixels that have, for example, more than 2 pixels clear in the kernel region with a threshold of 2.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

dilate

dilate(size: int, threshold: int | None = None, mask: Image | None = None) -> Image

Adds pixels to the edges of segmented areas.

This method works by convolving a kernel of ((size*2)+1)x((size*2)+1) pixels across the image and setting the center pixel of the kernel if the sum of the neighbour pixels set is greater than threshold.

This method works like the standard dilate method if threshold is not set. If threshold is set then you can specify dilate to only dilate pixels that have, for example, more than 2 pixels set in the kernel region with a threshold of 2.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

open

open(size: int, threshold: int | None = None, mask: Image | None = None) -> Image

Performs erosion and dilation on an image in order. Please see Image.erode() and Image.dilate() for more information.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

close

close(size: int, threshold: int | None = None, mask: Image | None = None) -> Image

Performs dilation and erosion on an image in order. Please see Image.dilate() and Image.erode() for more information.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

top_hat

top_hat(size: int, threshold: int | None = None, mask: Image | None = None) -> Image

Returns the image difference of the image and Image.open()’ed image.

Not supported on compressed images or bayer/yuv images.

black_hat

black_hat(size: int, threshold: int | None = None, mask: Image | None = None) -> Image

Returns the image difference of the image and Image.close()’ed image.

Not supported on compressed images or bayer/yuv images.

mean

mean(size: int, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Standard mean blurring filter using a box filter.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

If you’d like to adaptive threshold the image on the output of the filter you can pass threshold=True which will enable adaptive thresholding of the image which sets pixels to one or zero based on a pixel’s brightness in relation to the brightness of the kernel of pixels around them. A negative offset value sets more pixels to 1 as you make it more negative while a positive value only sets the sharpest contrast changes to 1. Set invert to invert the binary image resulting output.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

median

median(size: int, percentile: float | None = 0.5, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Runs the median filter on the image. The median filter is the best filter for smoothing surfaces while preserving edges but it is very slow.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

percentile controls the percentile of the value used in the kernel. By default each pixel is replaced with the 50th percentile (center) of its neighbors. You can set this to 0 for a min filter, 0.25 for a lower quartile filter, 0.75 for an upper quartile filter, and 1.0 for a max filter.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

mode

mode(size: int, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = Nonee) -> Image

Runs the mode filter on the image by replacing each pixel with the mode of their neighbors. This method works great on grayscale images. However, on RGB images it creates a lot of artifacts on edges because of the non-linear nature of the operation.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

midpoint

midpoint(size: int, bias: float | None = 0.5, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Runs the midpoint filter on the image. This filter finds the midpoint ((max-min)/2) of each pixel neighborhood in the image.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

bias controls the min/max mixing. 0 for min filtering only, 1.0 for max filtering only. By using the bias you can min/max filter the image.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

morph

morph(size: int, kernel: list, mul: float | None = 1.0, add: float | None = 0.0, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Convolves the image by a filter kernel. This allows you to do general purpose convolutions on an image.

size controls the size of the kernel which must be ((size*2)+1)x((size*2)+1) elements big.

kernel is the kernel to convolve the image by. The kernel can either be a 1D tuple or list or a 2D tuple or list. For 1D kernels the tuple/list must be ((size*2)+1)x((size*2)+1) elements big. For 2D tuples/lists each row must be ((size*2)+1) elements big and there must be ((size*2)+1) rows.

mul allows you to do a global contrast adjustment. It’s value should be greater than 0.0. The default value is 1.0 which does nothing.

add allows you to do a global brightness adjustment. It’s value should be between 0.0 and 1.0. The default value is 0.0 which does nothing.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

gaussian

gaussian(size: int, unsharp: bool | None = False, mul: float | None = 1.0, add: float | None = 0.0, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Convolves the image by a smoothing gaussian kernel.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

If unsharp is set to the True then instead of doing just a gaussian filtering operation this method will perform an unsharp mask operation which improves image sharpness on edges.

mul allows you to do a global contrast adjustment. It’s value should be greater than 0.0. The default value is 1.0 which does nothing.

add allows you to do a global brightness adjustment. It’s value should be between 0.0 and 1.0. The default value is 0.0 which does nothing.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

laplacian

laplacian(size: int, sharpen: bool | None = False, mul: float | None = 1.0, add: float | None = 0.0, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Convolves the image by a edge detecting laplacian kernel.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

If sharpen is set to the True then instead of just outputting an unthresholded edge detection image this method will instead sharpen the image. Increase the kernel size then to increase the image sharpness.

mul allows you to do a global contrast adjustment. It’s value should be greater than 0.0. The default value is 1.0 which does nothing.

add allows you to do a global brightness adjustment. It’s value should be between 0.0 and 1.0. The default value is 0.0 which does nothing.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

bilateral

bilateral(size: int, color_sigma: float | None = 0.1, space_sigma: float | None = 1.0, threshold: bool | None = False, offset: int | None = 0, invert: bool | None = False, mask: Image | None = None) -> Image

Convolves the image by a bilateral filter. The bilateral filter smooths the image while keeping edges in the image.

size is the kernel size. Use 1 (3x3 kernel), 2 (5x5 kernel), etc.

color_sigma controls how closely colors are matched using the bilateral filter. Increase this to increase color blurring.

space_sigma controls how closely pixels space-wise are blurred with each other. Increase this to increase pixel blurring.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer/yuv images.

Geometric Methods

linpolar

linpolar(reverse: bool = False) -> Image

Re-project’s and image from cartessian coordinates to linear polar coordinates.

Set reverse=True to re-project in the opposite direction.

Linear polar re-projection turns rotation of an image into x-translation.

Not supported on compressed images or bayer images.

This method is not available on the OpenMV Cam M4.

logpolar

logpolar(reverse: bool = False) -> Image

Re-project’s and image from cartessian coordinates to log polar coordinates.

Set reverse=True to re-project in the opposite direction.

Log polar re-projection turns rotation of an image into x-translation and scaling/zooming into y-translation.

Not supported on compressed images or bayer images.

This method is not available on the OpenMV Cam M4.

lens_corr

lens_corr(strength: float = 1.8, zoom: float = 1.0, x_corr: float = 0.0, y_corr: float = 0.0) -> Image

Performs lens correction to un-fisheye the image due to the lens distortion.

strength is a float defining how much to un-fisheye the image. Try 1.8 out by default and then increase or decrease from there until the image looks good.

zoom is the amount to zoom in on the image by. 1.0 by default.

x_corr floating point pixel offset from center. Can be negative or positive.

y_corr floating point pixel offset from center. Can be negative or positive.

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

rotation_corr

rotation_corr(x_rotation=0.0, y_rotation=0.0, z_rotation=0.0, x_translation=0.0, y_translation=0.0, zoom=1.0, fov=60.0, corners: List[Tuple[int, int]] | None = None) -> Image

Corrects perspective issues in the image by doing a 3D rotation of the frame buffer.

x_rotation is the number of degrees to rotation the image in the frame buffer around the x axis (i.e. this spins the image up and down).

y_rotation is the number of degrees to rotation the image in the frame buffer around the y axis (i.e. this spins the image left and right).

z_rotation is the number of degrees to rotation the image in the frame buffer around the z axis (i.e. this spins the image in place).

x_translation is the number of units to move the image to the left or right after rotation. Because this translation is applied in 3D space the units aren’t pixels…

y_translation is the number of units to move the image to the up or down after rotation. Because this translation is applied in 3D space the units aren’t pixels…

zoom is the amount to zoom in on the image by. 1.0 by default.

fov is the field-of-view to use internally when doing 2D->3D projection before rotating the image in 3D space. As this value approaches 0 the image is placed at infinity away from the viewport. As this value approaches 180 the image is placed within the viewport. Typically, you should not change this value but you can modify it to change the 2D->3D mapping effect.

corners is a list of four (x,y) tuples representing four corners used to create a 4-point correspondence homography that will map the first corner to (0, 0), the second corner to (image_width-1, 0), the third corner to (image_width-1, image_height-1), and the fourth corner to (0, image_height-1). The 3D rotation is then applied after the image is re-mapped. This argument lets you use rotation_corr to do things like birds-eye-view transforms. E.g:

top_tilt = 10 # if the difference between top/bottom_tilt become to large this method will stop working
bottom_tilt = 0

points = [(tilt, 0), (img.width()-tilt, 0), (img.width()-1-bottom_tilt, img.height()-1), (bottom_tilt, img.height()-1)]

img.rotation_corr(corners=points)

Returns the image object so you can call another method using . notation.

Not supported on compressed images or bayer images.

This method is not available on the OpenMV Cam M4.

Get Methods

get_similarity

get_similarity(image: Image, x: int | None = 0, y: int | None = 0, x_scale: float = 1.0, y_scale: float = 1.0, roi: Tuple[int, int, int, int] | None = None, rgb_channel: int = -1, alpha: int = 256, color_palette=None, alpha_palette=None, hint: int = 0, dssim: bool = False) -> Similarity

Computes the similarity between two images. The similarity is computed by using the structural similarity index (SSIM). The SSIM is a metric that compares the structural similarity between two images. The SSIM is a value between -1 and 1. A value of 1 means the images are identical, a value of 0 means the images are not similar, and a value of -1 means the images are perfectly the opposite of each other. Typically, if you want to check if two images are different you should look to see how negative the SSIM value is.

image is the image to compare to.

You may also pass a path instead of an image object for this method to automatically load the image from disk and use it in one step. E.g. get_similarity("test.jpg").

x is the x offset to start comparing the image at.

y is the y offset to start comparing the image at.

roi is the region-of-interest rectangle tuple (x, y, w, h) of the source image. This allows you to extract just the pixels in the ROI.

rgb_channel is the RGB channel (0=R, G=1, B=2) to extract from an RGB565 image (if passed). For example, if you pass rgb_channel=1 this will extract the green channel of the source RGB565 image and apply that in grayscale.

color_palette if not None can be image.PALETTE_RAINBOW, image.PALETTE_IRONBOW, or a 256 pixel in total RGB565 image to use as a color lookup table on the grayscale value of whatever the source image is. This is applied after rgb_channel extraction if used.

alpha_palette if not None can be a 256 pixel in total GRAYSCALE image to use as a alpha palette which modulates the alpha value of the source image being at a pixel pixel level allowing you to precisely control the alpha value of pixels based on their grayscale value. A pixel value of 255 in the alpha lookup table is opaque which anything less than 255 becomes more transparent until 0. This is applied after rgb_channel extraction if used.

hint can be a logical OR of the flags:

image.AREA: Use area scaling when downscaling versus the default of nearest neighbor.
image.BILINEAR: Use bilinear scaling versus the default of nearest neighbor scaling.
image.BICUBIC: Use bicubic scaling versus the default of nearest neighbor scaling.
image.CENTER: Center the image being operated on the display. This is applied after scaling.
image.HMIRROR: Horizontally mirror the source image.
image.VFLIP: Vertically flip the source image.
image.TRANSPOSE: Transpose the source image (swap x/y).
image.EXTRACT_RGB_CHANNEL_FIRST: Do rgb_channel extraction before scaling.
image.APPLY_COLOR_PALETTE_FIRST: Apply color palette before scaling.
image.SCALE_ASPECT_KEEP: Scale the source image being operated to fit inside the destination.
image.SCALE_ASPECT_EXPAND: Scale the source image being operated to fill the destination (results in cropping)
image.SCALE_ASPECT_IGNORE: Scale the source image being operated to fill the destination (results in stretching).
image.ROTATE_90: Rotate the source image by 90 degrees (this is just VFLIP | TRANSPOSE).
image.ROTATE_180: Rotate the source image by 180 degrees (this is just HMIRROR | VFLIP).
image.ROTATE_270: Rotate the source image by 270 degrees (this is just HMIRROR | TRANSPOSE).
image.BLACK_BACKGROUND: Assume the destination image being operated on is black speeding up blending.

dssim if true will compute the structural disimilarity index (DSSIM) instead of the SSIM. A value of 0 means the images are identical. A value of 1 means the images are completely different.

Returns a image.Similarity object.

get_histogram

get_histogram(thresholds: List[Tuple[int, int]] | None = None, invert=False, roi: Tuple[int, int, int, int] | None = None, bins=256, l_bins=256, a_bins=256, b_bins=256, difference: Image | None = None) -> histogram

Computes the normalized histogram on all color channels for an roi and returns a image.histogram object. Please see the image.histogram object for more information. You can also invoke this method by using Image.get_hist() or Image.histogram(). If you pass a list of thresholds then the histogram information will only be computed from pixels within the threshold list.

thresholds must be a list of tuples [(lo, hi), (lo, hi), ..., (lo, hi)] defining the ranges of color you want to track. For grayscale images each tuple needs to contain two values - a min grayscale value and a max grayscale value. Only pixel regions that fall between these thresholds will be considered. For RGB565 images each tuple needs to have six values (l_lo, l_hi, a_lo, a_hi, b_lo, b_hi) - which are minimums and maximums for the LAB L, A, and B channels respectively. For easy usage this function will automatically fix swapped min and max values. Additionally, if a tuple is larger than six values the rest are ignored. Conversely, if the tuple is too short the rest of the thresholds are assumed to be at maximum range.

Note

You may also determine color thresholds by going into Tools->Machine Vision->Threshold Editor in OpenMV IDE and selecting thresholds from the GUI slider window.

invert inverts the thresholding operation such that instead of matching pixels inside of some known color bounds pixels are matched that are outside of the known color bounds.

Unless you need to do something advanced with color statistics just use the Image.get_statistics() method instead of this method for looking at pixel areas in an image.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

bins and others are the number of bins to use for the histogram channels. For grayscale images use bins and for RGB565 images use the others for each channel. The bin counts must be greater than 2 for each channel. Additionally, it makes no sense to set the bin count larger than the number of unique pixel values for each channel. By default, the histogram will have the maximum number of bins per channel.

difference may be set to an image object to cause this method to operate on the difference image between the current image and the difference image object. This saves having to use a separate buffer.

Not supported on compressed images or bayer images.

get_statistics

get_statistics(thresholds: List[Tuple[int, int]] | None = None, invert=False, roi: Tuple[int, int, int, int] | None = None, bins=256, l_bins=256, a_bins=256, b_bins=256, difference: Image | None = None) -> statistics

Computes the mean, median, mode, standard deviation, min, max, lower quartile, and upper quartile for all color channels for an roi and returns a image.statistics object. Please see the image.statistics object for more information. You can also invoke this method by using Image.get_stats() or Image.statistics(). If you pass a list of thresholds then the histogram information will only be computed from pixels within the threshold list.

Note

You may also determine color thresholds by going into Tools->Machine Vision->Threshold Editor in OpenMV IDE and selecting thresholds from the GUI slider window.

invert inverts the thresholding operation such that instead of matching pixels inside of some known color bounds pixels are matched that are outside of the known color bounds.

You’ll want to use this method any time you need to get information about the values of an area of pixels in an image. For example, after if you’re trying to detect motion using frame differencing you’ll want to use this method to determine a change in the color channels of the image to trigger your motion detection threshold.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

Not supported on compressed images or bayer images.

get_regression

get_regression(thresholds: List[Tuple[int, int]], invert=False, roi: Tuple[int, int, int, int] | None = None, x_stride=2, y_stride=1, area_threshold=10, pixels_threshold=10, robust=False) -> line

Computes a linear regression on all the thresholded pixels in the image. The linear regression is computed using least-squares normally which is fast but cannot handle any outliers. If robust is True then the Theil–Sen linear regression is used instead which computes the median of all slopes between all thresholded pixels in the image. This is an N^2 operation which may drops your FPS down to below 5 even on an 80x60 image if too many pixels are set after thresholding. However, as long as the number of pixels set after thresholding remains low the linear regression will be valid even in the case of up to 30% of the thresholded pixels being outliers (e.g. it’s robust).

This method returns a image.line object. See this blog post on how to use the line object easily: https://openmv.io/blogs/news/linear-regression-line-following

Note

You may also determine color thresholds by going into Tools->Machine Vision->Threshold Editor in OpenMV IDE and selecting thresholds from the GUI slider window.

invert inverts the thresholding operation such that instead of matching pixels inside of some known color bounds pixels are matched that are outside of the known color bounds.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

x_stride is the number of x pixels to skip over when evaluating the image.

y_stride is the number of y pixels to skip over when evaluating the image.

If the regression’s bounding box area is less than area_threshold then None is returned.

If the regression’s pixel count is less than pixels_threshold then None is returned.

Not supported on compressed images or bayer images.

Detection Methods

find_blobs

find_blobs(thresholds: List[Tuple[int, int]], invert=False, roi: Tuple[int, int, int, int] | None = None, x_stride=2, y_stride=1, area_threshold=10, pixels_threshold=10, merge=False, margin=0, threshold_cb=None, merge_cb=None, x_hist_bins_max=0, y_hist_bins_max=0) -> List[blob]

Finds all blobs (connected pixel regions that pass a threshold test) in the image and returns a list of image.blob objects which describe each blob. Please see the image.blob object more more information.

thresholds must be a list of tuples [(lo, hi), (lo, hi), ..., (lo, hi)] defining the ranges of color you want to track. You may pass up to 32 threshold tuples in one call. For grayscale images each tuple needs to contain two values - a min grayscale value and a max grayscale value. Only pixel regions that fall between these thresholds will be considered. For RGB565 images each tuple needs to have six values (l_lo, l_hi, a_lo, a_hi, b_lo, b_hi) - which are minimums and maximums for the LAB L, A, and B channels respectively. For easy usage this function will automatically fix swapped min and max values. Additionally, if a tuple is larger than six values the rest are ignored. Conversely, if the tuple is too short the rest of the thresholds are assumed to be at maximum range.

Note

You may also determine color thresholds by going into Tools->Machine Vision->Threshold Editor in OpenMV IDE and selecting thresholds from the GUI slider window.

invert inverts the thresholding operation such that instead of matching pixels inside of some known color bounds pixels are matched that are outside of the known color bounds.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

x_stride is the number of x pixels to skip when searching for a blob. Once a blob is found the line fill algorithm will be pixel accurate. Increase x_stride to speed up finding blobs if blobs are known to be large.

y_stride is the number of y pixels to skip when searching for a blob. Once a blob is found the line fill algorithm will be pixel accurate. Increase y_stride to speed up finding blobs if blobs are known to be large.

If a blob’s bounding box area is less than area_threshold it is filtered out.

If a blob’s pixel count is less than pixels_threshold it is filtered out.

merge if True merges all not filtered out blobs whose bounding rectangles intersect each other. margin can be used to increase or decrease the size of the bounding rectangles for blobs during the intersection test. For example, with a margin of 1 blobs whose bounding rectangles are 1 pixel away from each other will be merged.

Merging blobs allows you to implement color code tracking. Each blob object has a code value which is a bit vector made up of 1s for each color threshold. For example, if you pass Image.find_blobs two color thresholds then the first threshold has a code of 1 and the second 2 (a third threshold would be 4 and a fourth would be 8 and so on). Merged blobs logically OR all their codes together so that you know what colors produced them. This allows you to then track two colors if you get a blob object back with two colors then you know it might be a color code.

You might also want to merge blobs if you are using tight color bounds which do not fully track all the pixels of an object you are trying to follow.

Finally, if you want to merge blobs, but, don’t want two color thresholds to be merged then just call Image.find_blobs twice with separate thresholds so that blobs aren’t merged.

threshold_cb may be set to the function to call on every blob after its been thresholded to filter it from the list of blobs to be merged. The call back function will receive one argument - the blob object to be filtered. The call back then must return True to keep the blob and False to filter it.

merge_cb may be set to the function to call on every two blobs about to be merged to prevent or allow the merge. The call back function will receive two arguments - the two blob objects to be merged. The call back then must return True to merge the blobs or False to prevent merging the blobs.

x_hist_bins_max if set to non-zero populates a histogram buffer in each blob object with an x_histogram projection of all columns in the object. This value then sets the number of bins for that projection.

y_hist_bins_max if set to non-zero populates a histogram buffer in each blob object with an y_histogram projection of all rows in the object. This value then sets the number of bins for that projection.

Not supported on compressed images or bayer images.

find_lines

find_lines(roi: Tuple[int, int, int, int] | None = None, x_stride=2, y_stride=1, threshold=1000, theta_margin=25, rho_margin=25) -> List[line]

Finds all infinite lines in the image using the hough transform. Returns a list of image.line objects.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

x_stride is the number of x pixels to skip when doing the hough transform. Only increase this if lines you are searching for are large and bulky.

y_stride is the number of y pixels to skip when doing the hough transform. Only increase this if lines you are searching for are large and bulky.

threshold controls what lines are detected from the hough transform. Only lines with a magnitude greater than or equal to threshold are returned. The right value of threshold for your application is image dependent. Note that the magnitude of a line is the sum of all sobel filter magnitudes of pixels that make up that line.

theta_margin controls the merging of detected lines. Lines which are theta_margin degrees apart and rho_margin rho apart are merged.

rho_margin controls the merging of detected lines. Lines which are theta_margin degrees apart and rho_margin rho apart are merged.

This method working by running the sobel filter over the image and taking the magnitude and gradient responses from the sobel filter to feed a hough transform. It does not require any preprocessing on the image first. However, my cleaning up the image using filtering you may get more stable results.

Not supported on compressed images or bayer images.

This method is not available on the OpenMV Cam M4.

find_line_segments

find_line_segments(roi: Tuple[int, int, int, int] | None = None, merge_distance=0, max_theta_difference=15) -> List[line]

Finds line segments in the image using the hough transform. Returns a list of image.line objects .

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

merge_distance specifies the maximum number of pixels two line segments can be separated by each other (at any point on one line) to be merged.

max_theta_difference is the maximum theta difference in degrees two line segments that are merge_distance apart to be merged.

This method uses the LSD library (also used by OpenCV) to find line segments in the image. It’s somewhat slow but very accurate and lines don’t jump around.

This method is not available on the OpenMV Cam M4.

find_circles

find_circles(roi: Tuple[int, int, int, int] | None = None, x_stride=2, y_stride=1, threshold=2000, x_margin=10, y_margin=10, r_margin=10, r_min=2, r_max: int | None = None, r_step=2) -> List[circle]

Finds circles in the image using the hough transform. Returns a list of image.circle objects.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

x_stride is the number of x pixels to skip when doing the hough transform. Only increase this if circles you are searching for are large and bulky.

y_stride is the number of y pixels to skip when doing the hough transform. Only increase this if circles you are searching for are large and bulky.

threshold controls what circles are detected from the hough transform. Only circles with a magnitude greater than or equal to threshold are returned. The right value of threshold for your application is image dependent. Note that the magnitude of a circle is the sum of all sobel filter magnitudes of pixels that make up that circle.

x_margin controls the merging of detected circles. Circles which are x_margin, y_margin, and r_margin pixels apart are merged.

y_margin controls the merging of detected circles. Circles which are x_margin, y_margin, and r_margin pixels apart are merged.

r_margin controls the merging of detected circles. Circles which are x_margin, y_margin, and r_margin pixels apart are merged.

r_min controls the minimum circle radius detected. Increase this to speed up the algorithm. Defaults to 2.

r_max controls the maximum circle radius detected. Decrease this to speed up the algorithm. Defaults to min(roi.w/2, roi.h/2).

r_step controls how to step the radius detection by. Defaults to 2.

This method is not available on the OpenMV Cam M4.

find_rects

find_rects(roi: Tuple[int, int, int, int] | None = None, threshold=10000) -> List[rect]

Find rectangles in the image using the same quad detection algorithm used to find apriltags. Works best of rectangles that have good contrast against the background. The apriltag quad detection algorithm can handle any scale/rotation/shear on rectangles. Returns a list of image.rect objects.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

Rectangles which have an edge magnitude (which is computed by sliding the sobel operator across all pixels on the edges of the rectangle and summing their values) less than threshold are filtered out of the returned list. The correct value of threshold is depended on your application/scene.

This method is not available on the OpenMV Cam M4.

find_qrcodes

find_qrcodes(roi: Tuple[int, int, int, int] | None = None) -> List[qrcode]

Finds all qrcodes within the roi and returns a list of image.qrcode objects. Please see the image.qrcode object for more information.

QR Codes need to be relatively flat in the image for this method to work. You can achieve a flatter image that is not effected by lens distortion by either using the sensor.set_windowing() function to zoom in the on the center of the lens, Image.lens_corr() to undo lens barrel distortion, or by just changing out the lens for something with a narrower fields of view. There are machine vision lenses available which do not cause barrel distortion but they are much more expensive to than the standard lenses supplied by OpenMV.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

This method is not available on the OpenMV Cam M4.

find_apriltags

find_apriltags(roi: Tuple[int, int, int, int] | None = None, families=TAG36H11, fx=0.0, fy=0.0, cx: int | None = None, cy: int | None = None) -> List[apriltag]

Finds all apriltags within the roi and returns a list of image.apriltag objects. Please see the image.apriltag object for more information.

Unlike QR Codes, AprilTags can be detected at much farther distances, worse lighting, in warped images, etc. AprilTags are robust too all kinds of image distortion issues that QR Codes are not to. That said, AprilTags can only encode a numeric ID as their payload.

AprilTags can also be used for localization purposes. Each image.apriltag object returns its translation and rotation from the camera. The units of the translation are determined by fx, fy, cx, and cy which are the focal lengths and center points of the image in the X and Y directions respectively.

Note

To create AprilTags use the tag generator tool built-in to OpenMV IDE. The tag generator can create printable 8.5”x11” AprilTags.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

families is bitmask of tag families to decode. It is the logical OR of:

image.TAG16H5
image.TAG25H7
image.TAG25H9
image.TAG36H10
image.TAG36H11
image.ARTOOLKIT

By default it is just image.TAG36H11 which is the best tag family to use. Note that Image.find_apriltags() slows down per enabled tag family.

fx is the camera X focal length in pixels. For the standard OpenMV Cam this is (2.8 / 3.984) * 656. Which is the lens focal length in mm, divided by the camera sensor length in the X direction multiplied by the number of camera sensor pixels in the X direction (for the OV7725 camera).

fx is the camera Y focal length in pixels. For the standard OpenMV Cam this is (2.8 / 2.952) * 488. Which is the lens focal length in mm, divided by the camera sensor length in the Y direction multiplied by the number of camera sensor pixels in the Y direction (for the OV7725 camera).

cx is the image center which is just image.width()/2. This is not roi.w()/2.

cy is the image center which is just image.height()/2. This is not roi.h()/2.

Not supported on compressed images.

This method is not available on the OpenMV Cam M4.

find_datamatrices

find_datamatrices(roi: Tuple[int, int, int, int] | None = None, effort=200) -> List[datamatrix]

Finds all datamatrices within the roi and returns a list of image.datamatrix objects. Please see the image.datamatrix object for more information.

Data Matrices need to be relatively flat in the image for this method to work. You can achieve a flatter image that is not effected by lens distortion by either using the sensor.set_windowing() function to zoom in the on the center of the lens, Image.lens_corr() to undo lens barrel distortion, or by just changing out the lens for something with a narrower fields of view. There are machine vision lenses available which do not cause barrel distortion but they are much more expensive to than the standard lenses supplied by OpenMV.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

effort controls how much time to spend trying to find data matrix matches. The default value of 200 should be good for all use-cases. However, you may increase the effort, at a cost of the frame rate, to increase detection. You may also lower the effort to increase the frame rate, but, at a cost of detections… note that when effort is set to below 160 or so you will not detect anything anymore. Also note that you can make effort as high as you like. But, any values above 240 or so do not result in much increase in the detection rate.

This method is not available on the OpenMV Cam M4.

find_barcodes

find_barcodes(roi: Tuple[int, int, int, int] | None = None) -> List[barcode]

Finds all 1D barcodes within the roi and returns a list of image.barcode objects. Please see the image.barcode object for more information.

For best results use a 640 by 40/80/160 window. The lower the vertical res the faster everything will run. Since bar codes are linear 1D images you just need a lot of resolution in one direction and just a little resolution in the other direction. Note that this function scans both horizontally and vertically so you can use a 40/80/160 by 480 window if you want. Finally, make sure to adjust your lens so that the bar code is positioned where the focal length produces the sharpest image. Blurry bar codes can’t be decoded.

This function supports all these 1D barcodes (basically all barcodes):

image.EAN2
image.EAN5
image.EAN8
image.UPCE
image.ISBN10
image.UPCA
image.EAN13
image.ISBN13
image.I25
image.DATABAR (RSS-14)
image.DATABAR_EXP (RSS-Expanded)
image.CODABAR
image.CODE39
image.PDF417
image.CODE93
image.CODE128

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

This method is not available on the OpenMV Cam M4.

find_displacement

find_displacement(template: Image, roi: Tuple[int, int, int, int] | None = None, template_roi: Tuple[int, int, int, int] | None = None, logpolar=False) -> List[displacement]

Find the translation offset of the this image from the template. This method can be used to do optical flow. This method returns a image.displacement object with the results of the displacement calculation using phase correlation.

roi is the region-of-interest rectangle (x, y, w, h) to work in. If not specified, it is equal to the image rectangle.

template_roi is the region-of-interest rectangle (x, y, w, h) to work in. If not specified, it is equal to the image rectangle.

roi and template roi must have the same w/h but may have any x/y location in the image. You may slide smaller rois around a larger image to get an optical flow gradient image…

Image.find_displacement() normally computes the x/y translation between two images. However, if you pass logpolar=True it will instead find rotation and scale changes between the two images. The same image.displacement object result encodes both possible responses.

Not supported on compressed images or bayer images.

Note

Please use this method on power-of-2 image sizes (e.g. sensor.B64X64).

Not supported on compressed images or bayer images.

This method is not available on the OpenMV Cam M4.

find_template

find_template(template: Image, threshold: float, roi: Tuple[int, int, int, int] | None = None, step=2, search=SEARCH_EX) -> Tuple[int, int, int, int]

Tries to find the first location in the image where template matches using Normalized Cross Correlation. Returns a bounding box tuple (x, y, w, h) for the matching location otherwise None.

template is a small image object that is matched against this image object. Note that both images must be grayscale.

threshold is floating point number (0.0-1.0) where a higher threshold prevents false positives while lowering the detection rate while a lower threshold does the opposite.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

step is the number of pixels to skip past while looking for the template. Skipping pixels considerably speeds the algorithm up. This only affects the algorithm in SERACH_EX mode.

search can be either image.SEARCH_DS or image.SEARCH_EX. image.SEARCH_DS searches for the template using as faster algorithm than image.SEARCH_EX but may not find the template if it’s near the edges of the image. image.SEARCH_EX does an exhaustive search for the image but can be much slower than image.SEARCH_DS.

Only works on grayscale images.

find_features

find_features(cascade, threshold=0.5, scale=1.5, roi: Tuple[int, int, int, int] | None = None) -> List[Tuple[int, int, int, int]]

This method searches the image for all areas that match the passed in Haar Cascade and returns a list of bounding box rectangles tuples (x, y, w, h) around those features. Returns an empty list if no features are found.

cascade is a Haar Cascade object. See image.HaarCascade() for more details.

threshold is a threshold (0.0-1.0) where a smaller value increase the detection rate while raising the false positive rate. Conversely, a higher value decreases the detection rate while lowering the false positive rate.

scale is a float that must be greater than 1.0. A higher scale factor will run faster but will have much poorer image matches. A good value is between 1.35 and 1.5.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

find_eye

find_eye(roi: Tuple[int, int, int, int]) -> Tuple[int, int]

Searches for the pupil in a region-of-interest (x, y, w, h) tuple around an eye. Returns a tuple with the (x, y) location of the pupil in the image. Returns (0,0) if no pupils are found.

To use this function first use Image.find_features() with the frontalface HaarCascade to find someone’s face. Then use Image.find_features() with the eye HaarCascade to find the eyes on the face. Finally, call this method on the eye ROI returned by Image.find_features() to get the pupil coordinates.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

Only works on grayscale images.

find_lbp

find_lbp(roi: Tuple[int, int, int, int])

Extracts LBP (local-binary-patterns) keypoints from the region-of-interest (x, y, w, h) tuple. You can then use then use the image.match_descriptor() function to compare two sets of keypoints to get the matching distance.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

Only works on grayscale images.

find_keypoints

find_keypoints(roi: Tuple[int, int, int, int] | None = None, threshold=20, normalized=False, scale_factor=1.5, max_keypoints=100, corner_detector=CORNER_AGAST)

Extracts ORB keypoints from the region-of-interest (x, y, w, h) tuple. You can then use then use the image.match_descriptor() function to compare two sets of keypoints to get the matching areas. Returns None if no keypoints were found.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

threshold is a number (between 0 - 255) which controls the number of extracted corners. For the default AGAST corner detector this should be around 20. FOr the FAST corner detector this should be around 60-80. The lower the threshold the more extracted corners you get.

normalized is a boolean value that if True turns off extracting keypoints at multiple resolutions. Set this to true if you don’t care about dealing with scaling issues and want the algorithm to run faster.

scale_factor is a float that must be greater than 1.0. A higher scale factor will run faster but will have much poorer image matches. A good value is between 1.35 and 1.5.

max_keypoints is the maximum number of keypoints a keypoint object may hold. If keypoint objects are too big and causing out of RAM issues then decrease this value.

corner_detector is the corner detector algorithm to use which extracts keypoints from the image. It can be either image.CORNER_FAST or image.CORNER_AGAST. The FAST corner detector is faster but much less accurate.

Only works on grayscale images.

find_edges

find_edges(edge_type, threshold=(100, 200))

Turns the image to black and white leaving only the edges as white pixels.

image.EDGE_SIMPLE - Simple thresholded high pass filter algorithm.
image.EDGE_CANNY - Canny edge detection algorithm.

threshold is a two valued tuple containing a low threshold and high threshold. You can control the quality of edges by adjusting these values. It defaults to (100, 200).

Only works on grayscale images.

find_hog

find_hog(roi: Tuple[int, int, int, int] | None = None, size=8)

Replaces the pixels in the ROI with HOG (histogram of orientated graidients) lines.

roi is the region-of-interest rectangle tuple (x, y, w, h). If not specified, it is equal to the image rectangle. Only pixels within the roi are operated on.

Only works on grayscale images.

This method is not available on the OpenMV Cam M4.

stero_disparity

stero_disparity(reversed: bool = False, max_disparity: int = 64, threshold: int = 64)

Takes a double wide grayscale image that contains the output of two camera sensors side-by-side and replaces one of the images in the double wide image with the stero-disparity image where each pixel represents depth. E.g. if you have two 320x240 cameras then this method takes a 640x240 image.

reversed By default the left image is compared to the right image and the right image is then replaced. Pass true to compare the right image to the left image and replace the left image.

Note

The algorithm only works comparing left->right or right->left. If your camrea setup does not match this then you will get useless results.

max_disparity is the maximum distance to search for a matching pixel block using the sum-of-absolute differences algorithm. Larger values take exponentially longer to search with but result in higher quality images. Lower values make the algorithm run faster but may result in nonsense output.

threshold if the sum-of-absolute differences between two blocks is less than or equal to this threshold they are considered to be matching.

This method is only available on the Arduino Portenta.

Note

Even with our best SIMD effort this algorithm is not real-time on the Cortex-M7 processor. This is just a toy example algorithm showing off stero-disparity.

Constants

image.BINARY

image.BINARY: int

BINARY (bitmap) pixel format. Each pixel is 1-bit.

image.GRAYSCALE

image.GRAYSCALE: int

GRAYSCALE pixel format. Each pixel is 8-bits, 1-byte.

image.RGB565

image.RGB565: int

RGB565 pixel format. Each pixel is 16-bits, 2-bytes. 5-bits are used for red, 6-bits are used for green, and 5-bits are used for blue.

image.BAYER

image.BAYER: int

RAW BAYER image pixel format. If you try to make the frame size too big to fit in the frame buffer your OpenMV Cam will set the pixel format to BAYER so that you can capture images but no image processing methods will be operational.

image.YUV422

image.YUV422: int

A pixel format that is very easy to jpeg compress. Each pixel is stored as a grayscale 8-bit Y value followed by alternating 8-bit U/V color values that are shared between two Y values (8-bits Y1, 8-bits U, 8-bits Y2, 8-bits V, etc.). Only some image processing methods work with YUV422.

image.JPEG

image.JPEG: int

A JPEG image.

image.PNG

image.PNG: int

A PNG image.

image.PALETTE_RAINBOW

image.PALETTE_RAINBOW: int

Default OpenMV Cam color palette for thermal images using a smooth color wheel.

image.PALETTE_IRONBOW

image.PALETTE_IRONBOW: int

Makes images look like the FLIR Lepton thermal images using a very non-linear color palette.

image.AREA

image.AREA: int

Use area scaling when downscaling an image (Nearest Neighbor is used for upscaling).

You should use area scaling when downscaling for the highest visual quality.

image.BILINEAR

image.BILINEAR: int

Use bilinear scaling when upscaling an image. This produces a good quality scaled image output and is fast.

When downscaling an image this method will subsample the input image to produce the downscaled image. Use image.AREA for the highest quality downscaling if speed is not an issue.

image.BICUBIC

image.BICUBIC: int

Use bicubic scaling when upscaling an image. This produces a high quality scaled image output, but is slow.

When downscaling an image this method will subsample the input image to produce the downscaled image. Use image.AREA for the highest quality downscaling if speed is not an issue.

image.VFLIP

image.VFLIP: int

Vertically flip the image being drawn by draw_image.

image.HMIRROR

image.HMIRROR: int

Horizontally mirror the image being drawn by draw_image.

image.TRANSPOSE

image.TRANSPOSE: int

Transpose (swap x/y) the image being draw by draw_image.

image.CENTER

image.CENTER: int

Center the image being drawn to the center of the image/canvas it’s being drawn on. Any x/y offsets passed will move the image being drawn from the center by that amount.

image.EXTRACT_RGB_CHANNEL_FIRST

image.EXTRACT_RGB_CHANNEL_FIRST: int

When extracting an RGB channel from an RGB image using draw_image extract the channel first before scaling versus afterwards to prevent any artifacts.

image.APPLY_COLOR_PALETTE_FIRST

image.APPLY_COLOR_PALETTE_FIRST: int

When applying a color lookup table to an image using draw_image apply the color look table first before scaling versus afterwards to prevent any artifacts.

image.SCALE_ASPECT_KEEP

image.SCALE_ASPECT_KEEP: int

Scale the image being drawn to fit inside of the image/canvas being drawn on while maintaining the aspect ratio. Unless the image aspect ratios match the image being drawn will not completely cover the image/canvas being drawn on. Any x_scale/y_scale values passed will additionally scale the scaled image.

image.SCALE_ASPECT_EXPAND

image.SCALE_ASPECT_EXPAND: int

Scale the image being drawn to fill image/canvas being drawn on while maintaining the aspect ratio. Unless the image aspect ratios match the image being drawn will be cropped. Any x_scale/y_scale values passed will additionally scale the scaled image.

image.SCALE_ASPECT_IGNORE

image.SCALE_ASPECT_IGNORE: int

Scale the image being drawn to fill the image/canvas being drawn on. This does not maintain the aspect ratio of the image being drawn. Any x_scale/y_scale values passed will additionally scale the scaled image.

image.BLACK_BACKGROUND

image.BLACK_BACKGROUND: int

Speeds up draw_image when drawing on a black destination image when using alpha effects that require reading both source and destination pixels. This skips reading the destination pixel.

image.ROTATE_90

image.ROTATE_90: int

Rotate the image by 90 degrees (this is just image.VFLIP ORed with image.TRANSPOSE).

image.ROTATE_180

image.ROTATE_180: int

Rotate the image by 180 degrees (this is just image.HMIRROR ORed with image.VFLIP).

image.ROTATE_270

image.ROTATE_270: int

Rotate the image by 270 degrees (this is just image.HMIRROR ORed with image.TRANSPOSE).

image.JPEG_SUBSAMPLING_AUTO

image.JPEG_SUBSAMPLING_AUTO: int

Automatically select the best JPEG subsampling based on the image quality parameter.

image.JPEG_SUBSAMPLING_444

image.JPEG_SUBSAMPLING_444: int

Use 4:4:4 JPEG subsampling.

image.JPEG_SUBSAMPLING_422

image.JPEG_SUBSAMPLING_422: int

Use 4:2:2 JPEG subsampling. Note, you should force the jpeg subsampling to be 4:2:2 if you are streaming video via MJPEG for the best compatibility with third-party video players.