puzzle.utils.imageProcessing Namespace Reference

Functions
def	cropImage (image, template)
	Crop and resize a cover image, which has the same shape with the mask. More...
def	extract_region (img, verbose=False)
def	find_nonzero_mask (mask)
	Extract the coordinates of the non-zero elements (x,y style) from a mask image. More...
def	preprocess_real_puzzle (img, mask=None, areaThresholdLower=1000, areaThresholdUpper=8000, BoudingboxThresh=(30, 80), cannyThresh=(30, 60), WITH_AREA_THRESH=False, verbose=False, ret_thresh_mask=False)
	Preprocess the RGB image of a segmented puzzle piece in a circle area to obtain a mask. More...
def	preprocess_synthetic_puzzle (img, mask=None, areaThresholdLower=1000, areaThresholdUpper=8000, cannyThresh=(20, 80), verbose=False)
	Preprocess the RGB image of a segmented puzzle piece in a circle area to obtain a mask. More...
def	rotate_im (image, angle, mask=None)
	Generate rotated image and return working elements of it. More...
def	rotate_nd (image, angle, mask=None)
	Generate rotated image and return working elements of it. More...
def	white_balance (img)

Function Documentation

cropImage()

def puzzle.utils.imageProcessing.cropImage	(		image,
			template
	)

Crop and resize a cover image, which has the same shape with the mask.

Args: image: The source image. template: The mask image.

Returns: The resized image.

extract_region()

def puzzle.utils.imageProcessing.extract_region	(		img,
			verbose = False
	)

@brief Extract the regions from user's self-defined template image.

Args:
    img: The input image.
    verbose: The flag of whether to debug.

Returns:
    regions: The mask region list.

find_nonzero_mask()

def puzzle.utils.imageProcessing.find_nonzero_mask

(

mask

)

Extract the coordinates of the non-zero elements (x,y style) from a mask image.

Parameters

[in]

mask

Input mask image with 0 or 1

Returns

rcoords Coordinates of the non-zero elements (x,y) style

preprocess_real_puzzle()

def puzzle.utils.imageProcessing.preprocess_real_puzzle	(		img,
			mask = None,
			areaThresholdLower = 1000,
			areaThresholdUpper = 8000,
			BoudingboxThresh = (30,80),
			cannyThresh = (30, 60),
			WITH_AREA_THRESH = False,
			verbose = False,
			ret_thresh_mask = False
	)

Preprocess the RGB image of a segmented puzzle piece in a circle area to obtain a mask.

Note that the threshold is very important. It requires having prior knowledge.

Implementation was changed. It might actually work for multiple puzzles on the workmat based on flood filling operation. Will recover binary mark that gets individual pieces assuming none are touching. If touching, then grabbed as an entity.

Limited testing was done to confirm functionality of the revised implementation. It seems to work, but can sometimes fail to match though other processing may lead to a match.

Parameters

[in]	img	RGB image input.
[in]	mask	Mask image input.
[in]	areaThresholdLower	Lower threshold for area.
[in]	areaThresholdUpper	Upper threshold for area.
[in]	BoudingboxThresh	Size threshold of the bounding box area.
[in]	cannyThresh	Threshold for canny.
[in]	WITH_AREA_THRESH	Mainly for previous codes which have not set the BoudingboxThresh properly.
[in]	verbose	Debug verbosity lag.

Returns

seg_img_combined The mask region list.

preprocess_synthetic_puzzle()

def puzzle.utils.imageProcessing.preprocess_synthetic_puzzle	(		img,
			mask = None,
			areaThresholdLower = 1000,
			areaThresholdUpper = 8000,
			cannyThresh = (20, 80),
			verbose = False
	)

Preprocess the RGB image of a segmented puzzle piece in a circle area to obtain a mask.

Todo:

Maybe can combine preprocess_real_puzzle and preprocess_synthetic_puzzle together?

Parameters

[in]	img	RGB image input.
[in]	mask	Mask image input.
[in]	areaThresholdLower	The lower threshold of the area.
[in]	areaThresholdUpper	The upper threshold of the area.
[in]	cannyThresh	The threshold for canny.
[in]	verbose	The flag of whether to debug.

Returns

seg_img_combined Mask region list.

rotate_im()

def puzzle.utils.imageProcessing.rotate_im	(		image,
			angle,
			mask = None
	)

Generate rotated image and return working elements of it.

See code example. (clockwise)

Parameters

[in]	image	Input image.
[in]	angle	Rotation angle.
[in]	mask	Optional mask to rotate also.

Returns

rotResults A tuple of: rotated image (cropped) & rotated image (not cropped) & the rotation matrix & padding_left & padding_top.

rotate_nd()

def puzzle.utils.imageProcessing.rotate_nd	(		image,
			angle,
			mask = None
	)

Generate rotated image and return working elements of it.

This code is from the stackoverflow solution for an OpenCV non-cropping image rotation. Someone provided a scipiy ndimage version as a solution too.

The image and the mask should be the same dimension. No checking for that being the case. Outer scope should deal with ensuring match or dealing with mismatch.

Parameters

[in]	image	Input image.
[in]	angle	Rotation angle (in degrees).
[in]	mask	Optional mask to rotate also.

Returns

rotResults Either the rotate image or a tuple of rotated image and mask.

white_balance()

def puzzle.utils.imageProcessing.white_balance

(

img

)

@brief Change the white balance of the image.

Args:
    img: Input image.

Returns:
    result: The processed image with white balance.