Image Match return True when the images have different color.

[gzll]

lackey/lackey/TemplateMatchers.py

Line 24 in 7adadfa

method = cv2.TM_CCOEFF_NORMED

the app gui have some image buttons, when click them, their color will become yellow from blue.


the find() or exists() function will return the match info, when the button is blue, and the match image is yellow, this is incorrect.
but if inverse, the button is yellow, and the match image is blue, the function doesn't find anything. this is correct.
why?
I compare lackey with SikuliX1's source code (doFindMatch() in Finder2.java), but cannot get any idea.

[glitchassassin]

Lackey (and Sikuli, if I'm not mistaken) converts the images to greyscale for comparison to improve performance. If you're interested in checking the color of a point, you can try using the Location.getColor() method.

[glitchassassin]

For reference, the code snippet you linked is to the older naive matcher; the Pyramid matcher, based on the same algorithm Sikuli uses, is the one that converts to greyscale first:

lackey/lackey/TemplateMatchers.py

Lines 71 to 88 in 7adadfa

	class PyramidTemplateMatcher(object):
	""" Python wrapper for OpenCV's TemplateMatcher
	Uses a pyramid model to optimize matching speed
	"""
	def __init__(self, haystack):
	self.haystack = cv2.cvtColor(haystack, cv2.COLOR_BGR2GRAY) # Convert to grayscale
	self._iterations = 3 # Number of times to downsample
	def findBestMatch(self, needle, similarity):
	""" Finds the best match using a search pyramid to improve efficiency
	Pyramid implementation unashamedly stolen from https://github.com/stb-tester/stb-tester
	*Developer's Note - Despite the name, this method actually returns the **first** result
	with enough similarity, not the **best** result.*
	"""
	needle = cv2.cvtColor(needle, cv2.COLOR_BGR2GRAY) # Convert to grayscale

[gzll]

Got, Convert to grayscale for performance reason.
I test the scenario in lackey and SIKULIX1, but results is very different.
the scenario: the blue image button, compared with a blue image button screen cropped, and a yellow one.
lackey: if a blue crap, the confidence is 1.0, if yellow, it is something like 0.9995211958885193.
SIKULIX1(test with the "pattern setting"'s "Matching Preview" GUI): if blue, 0.99, if yellow, never matched, but when the confidence is down below 0.4, the matching is another yellow image button on the GUI.
so I think SIKULIX1 maybe still keeping the BGR channels for matching.
https://github.com/RaiMan/SikuliX1/blob/98ac9872e6d186c699cebcef1dc44e72dc17f87a/API/src/main/java/org/sikuli/script/FindInput2.java#L213
https://github.com/RaiMan/SikuliX1/blob/98ac9872e6d186c699cebcef1dc44e72dc17f87a/API/src/main/java/org/sikuli/script/Finder2.java#L605
https://github.com/RaiMan/SikuliX1/blob/98ac9872e6d186c699cebcef1dc44e72dc17f87a/API/src/main/java/org/sikuli/script/Finder2.java#L349

[glitchassassin]

At a casual glance, I think this is related to improvements noted in #136, where Sikuli now supports transparent images as masks. I'll do some research on this.