4

I have three images:

enter image description here

enter image description here

enter image description here

I've aligned them spatio-temporally to the best of my ability. I want to detect where the content-specific differences (the caption, the RT logo) are. Simple image difference (followed by thresholding, etc) doesn't work here for a couple of reasons:

  • the first image is quite blurry. Image difference will yield peaks at edges. I don't want that
  • the first image has undergone some sort of color transformation, so the difference contains a lot of information that I don't need

Here's the plain image difference between the first two images (using GIMP):

enter image description here

Any ideas?

mpenkov
  • 21,621
  • 10
  • 84
  • 126
  • 3
    Might be of help: http://stackoverflow.com/questions/2219185/duplicate-image-detection-algorithms – Blagovest Buyukliev Jun 21 '11 at 11:25
  • 1
    I agree it's not the approach to take, but this specific example doesn't entirely justify your assertion that diff+threshold is ruled out -- the similar areas are not identical, but the differences are *much* bigger. You probably could make do with that and bit of filtering... – walkytalky Jun 21 '11 at 12:09
  • @walkytalky: yeah, if i low-pass filter the difference before thresholding, i can get a better result. the problem is that it's hard to determine the threshold other than by empirical means – mpenkov Jun 21 '11 at 14:33

1 Answers1

2

You could partition each image into a grid of squares and compute local histograms for each square. Then you can compare the corresponding histograms from the three images. The histograms that differ most should identify the regions with the largest differences.

Blagovest Buyukliev
  • 42,498
  • 14
  • 94
  • 130