As related to the project I am currently involved in, I have come to realize that one thing that might help is for me to 'segment' my image. To that end, I have shown a very simple example below.

How do segmentation methods work in general, and what class of them would help getting me from the first image:

enter image description here

...to the second:

enter image description here

Please ignore the scaling that might seem to have occured here, but you get the general idea.

  • 1
    $\begingroup$ For your example, the most basic way of segmentation to get the result you show, is of course thresholding $\endgroup$
    – Geerten
    Commented Mar 16, 2012 at 15:55
  • $\begingroup$ @Geerten Yes this is actually how I generated the above image. :-) $\endgroup$
    – Spacey
    Commented Mar 16, 2012 at 16:54

1 Answer 1


One of the simplest method you can use is

"Region growing" scheme. Here the algorithm starts with a seed points, and keeps registering the pixel surrounded by it as long as the region as a whole has same property as before and after the inclusion of pixel. This is applied with reasonably good set of seed points recursively.

This property can be color variations, statistical properties such as mean and variance of intensity of the regions, or it can be texture. And the criteria for segmentation is homogeneity of the region.

The same principle but alternative implementation is region split and merge approach. Here a image is first divided into fixed parts - say 4 parts and corresponding scores of above properties is computed. Now the each sub region is split further, and we can check if any region is significantly different from others or same. If homongenety of sub parts is same essentially it is merged back. But if the element is realized to have different properties it now fall in different regions.

Read this for more details: http://vplab.iitm.ac.in/courses/CV_DIP/PDF/lect-Segmen.pdf

  • $\begingroup$ One of my favorite segmentation methods, works very well, and with the correct implementation, fast enough too. Using different distance measures can give you a very specific and smart segmentation method. $\endgroup$
    – Geerten
    Commented Mar 15, 2012 at 10:31
  • $\begingroup$ the pdf link is broken. Please can you provide other link? $\endgroup$ Commented Dec 9, 2016 at 13:12

Your Answer

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

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