6

Goal

I need to check if one image is a "sub set of another image". I am using following two functions, which seem to be working occasionally, but I am looking for some improvements to make it reliable. The functions, I am using are A and B below. Their goal is to check if image X is a subset of Y - or in other words check if Y image contains image X. ( The caller function takes care of which images to pass )

Note: The terminology of super-set, sub-set, union and intersection is adapted from set theory, if it helps to understand the question.

Caveat:

  • The images may not be "pixel same", but they might look exactly same to naked eye. Hence the functions A and B below are returning a number using getImageDifference , where 0 is a perfect absolute match, and non zero numbers are treated relatively by callee function based on what is being returned for other queries.

  • Only Pillow (PIL) and/or NumPy could be used

Functions

  • function A : if X is subset of Y, then the intersection of X and Y is X
  • function B : if x is subset of Y, then the union of X and Y is Y

Related articles :

https://wildclick.wordpress.com/2016/07/09/python-pil-pillow-numpy-intersect-images/ and https://wildclick.wordpress.com/2016/07/08/python-pil-pillow-numpy-add-images/

Existing Code :

Image Comparer

def getImageDifference(self, x, y):
        # Calculate the difference b/w two images, works
        return diff3
        pass

Function A and B

def A(self, x, y):
        '''
        check if image X is a subset of Y, using intersection
        :param x:
        :param y:
        :return:
        '''

        xx = Image.open(self.getImagePathByName(x)).convert(mode='L', dither=Image.NONE)
        yy = Image.open(self.getImagePathByName(y)).convert(mode='L', dither=Image.NONE)

        pixelThreshold = 200
        xx = np.array(xx)
        xx = np.where(xx > pixelThreshold, 255, 0)
        yy = np.array(yy)
        yy = np.where(yy > pixelThreshold, 255, 0)
        intersection = (xx + yy)

        # if xx is a subset of yy then the intersection of xx and yy, is xx
        xx = Image.open(self.getImagePathByName(x)).convert(mode='L', dither=Image.NONE)
        intersection = Image.fromarray(intersection.astype(np.uint8)).convert(mode='L', dither=Image.NONE)

        difference = self.getImageDifference(xx, intersection)
        return difference
        pass


def B(self, x, y):
        '''
        check if image X is a subset of Y, using union
        :param x:
        :param y:
        :return:
        '''

        xx = Image.open(self.getImagePathByName(x)).convert(mode='L', dither=Image.NONE)
        yy = Image.open(self.getImagePathByName(y)).convert(mode='L', dither=Image.NONE)

        pixelThreshold = 200
        xx = np.array(xx)
        xx = np.where(xx > pixelThreshold, 255, 0)
        yy = np.array(yy)
        yy = np.where(yy > pixelThreshold, 255, 0)
        union = (xx + yy) / 2
        union = np.where(union > pixelThreshold, 255, 0)

        # if xx is a subset of yy then the union of xx and yy, is yy
        yy = Image.open(self.getImagePathByName(y)).convert(mode='L', dither=Image.NONE)
        union = Image.fromarray(union.astype(np.uint8)).convert(mode='L', dither=Image.NONE)

        difference = self.getImageDifference(yy, union)
        return difference
        pass
ablaze
  • 722
  • 7
  • 30
  • test images: http://imgur.com/a/lFyav – ablaze Jul 23 '16 at 19:27
  • See if this helps : [`How can I check if one two-dimensional NumPy array contains a specific pattern of values inside it?`](http://stackoverflow.com/questions/32531377/how-can-i-check-if-one-two-dimensional-numpy-array-contains-a-specific-pattern-o). – Divakar Jul 23 '16 at 19:31
  • So, what’s the question again? I see goals and code, but no question? – Ahmed Fasih Jul 25 '16 at 05:57
  • As mentioned in "Goals" the current function is not reliable. The goal is only to determine if an image is a subset of another image, please refer caveats before answering. I've also provided test images in the first comment here – ablaze Jul 25 '16 at 07:28

0 Answers0