1
$\begingroup$

I'm looking for methods of computing some measure of structure in an image. (I do not have a rigorous definition of "structure" though........) Consider the 15 image blocks below:

Example of structures in images

Suppose for each pixel black = -1, white = 1 and grey = 0. Anything significantly far from 0 is "interesting".

The ones with significant white regions clearly have some structure to them. One way to identify them is by averaging the absolute values of the pixels then set some minimum threshold.

This works, but it does not take into consideration any spatial patterns. So it is unlikely to detect that the blocks along the bottom, and along right, also arguably have some sort of structure to them.

In either case, arguably the top left two blocks have a low structure measure i.e. they looks mostly noise.

Are there any general measures of structure in images which might help for examples like this? Edge detection, filters, etc..... Where to start?

Improve this question
$\endgroup$
4
  • $\begingroup$ It would be good to know what these images are and whether their size is fixed (?). By the way, if these are connectivity matrices, there are other ways of identifying structure in them. $\endgroup$
    – A_A
    Sep 25, 2018 at 9:39
  • $\begingroup$ Each block is a correlation matrix between pairs of time-varying variables. $\endgroup$
    – Brendan Hill
    Sep 26, 2018 at 9:41
  • $\begingroup$ Have you looked into community structure for weighted networks? Image processing techniques (texture analysis) are not likely to help you here. Would an answer along the "connectivity" lines suffice? $\endgroup$
    – A_A
    Sep 26, 2018 at 9:50
  • $\begingroup$ Do a fourier transform and remove frequencies below a threshold like 3? See what it looks like without the high frequency noise. $\endgroup$
    – ziggy jones
    Oct 1, 2018 at 14:23

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.