I have an image that includes object and background. However, the object appears some inhomogeneity region due to illumination. My work is that how to detect inhomogeneity region. Which is feature can represent it? Example: low inhomogeneity-high/low feature value....Thank you so much

enter image description here


enter image description here

  • $\begingroup$ This question has been cross-posted at StackOverflow: stackoverflow.com/questions/24941115/… $\endgroup$
    – John1024
    Jul 24, 2014 at 18:21
  • $\begingroup$ Searching Google for this image turned up several papers that showed how to segment it; e.g., Implicit Active Contours Driven by Local Binary Fitting Energy $\endgroup$
    – Emre
    Jul 24, 2014 at 18:37
  • $\begingroup$ @Emre: Your suggestion paper is not my goal. Your paper only considers gaussian filter as local kernel. However, they set sigma manually, while that parameter is very important. It indicates the region where big/small inhomogeneity $\endgroup$
    – John
    Jul 25, 2014 at 2:08
  • $\begingroup$ @John, Could you please review my answer? Mark it if fits you. $\endgroup$
    – Royi
    Jun 20 at 18:49

1 Answer 1


There are many properties of inhomogeneity:

  1. Local Variance / STD.
  2. Local Histogram.
  3. The Gradient Function
  4. Histogram of the Gradient.
  5. Mean versus the Median / Mode.
  • $\begingroup$ How to take local feature? Only convolution the image with mask. Example mask of 3x3 elements...[1 1 1;1 1 1;1 1 1] $\endgroup$
    – John
    Jul 25, 2014 at 11:54
  • $\begingroup$ Sliding Window operations. Not all of them can be done by Convolution (Or at least require some "Pre Processing"). $\endgroup$
    – Royi
    Jul 25, 2014 at 11:57
  • $\begingroup$ @John, could you please review my answer. If anything is missing let me know. Otherwise, could you please mark it? $\endgroup$
    – Royi
    Jan 28 at 7:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.