0

In the below attached images, I have to find the differences and point them out in the images labelled as "#-right.png". I had planned to use otsu thresholding and subtraction of images to find out the differences but the problem is the image sizes are not identical (check picture resolutions and the varying borders around the images) for this method to work.

Even if I wanted to use geometrical transformations, I would have to manually point out the points to which the images have to be cropped out (which is different for different images) etc. Is there any way (or any concept I'm unaware of) to make these images identical in resolution so that I can carry on with the subtraction to find the differences?

7-right image 301x407 px

7-left image 271x405 px

1-left image 265x260 px

1-right image 259x264 px

T S Aditya
  • 25
  • 8
  • As far as pre-processing is concerned, You can segment the pure white background using contour detection. Then you can use either [`SIFT, SURF or ORB`](https://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_feature2d/py_matcher/py_matcher.html) for feature matching – ZdaR Dec 19 '18 at 06:15

1 Answers1

4

Yes, find keypoints and match them, then diff.

For the first pair:

(1) find and match

enter image description here

(2) reproject

enter image description here

(3) diff

enter image description here

For the second pair:

(1) find and match

enter image description here

(2) reproject

enter image description here

(3) diff

enter image description here

Releated:

  1. How do I use the relationships between Flann matches to determine a sensible homography?

  2. CV - Extract differences between two images

Kinght 金
  • 17,681
  • 4
  • 60
  • 74