4

Given an image of tubular-like structures from which the edges have been extracted, I would like to fill/flood the regions inside the edges. It's worth to mention that the edges represent an edgeness measure/likelihood. The more blue it looks like, the more probable an edge is to be found. Any ideas how to fill the regions?

I applied a set of successive morphological closings with structural elements of different sizes (circles with sizes of: 5x5,7x7,9x9,11x11,13x13,15x15). However the results are not so nice yet. What it's important to note is that the width of the tubular structures is not uniform, and also the input image exhibits some holes that shall not be filled in.

An illustration of an input image is given on the following link: enter image description here

An illustration of the output after the multi-scale closing is given at: enter image description here

Any suggestions are appreciated :-)

Tin
  • 1,006
  • 1
  • 15
  • 27

1 Answers1

0

Have a try with the morphological watershed on the edgeness map.

This is the kind of segmentation you can get.

enter image description here

  • Yves Daoust, do you mind in giving an illustration how did you generate your image? How did you define your markers, for instance? On the other hand, with your solution I would need to somehow merge regions afterwards in such a way that the tubular structures are merged. – Tin Jun 08 '16 at 12:38
  • @tin No markers defined, this is a straight watershed. "I would need to somehow merge regions": from the given picture, it is quite impossible to guess what the expected output is. And what you expect is probably subjective. –  Jun 08 '16 at 12:44
  • Yves Daoust, do you mind in illustrating your solution with the corresponding matlab code? After running the watershed directly on the edge map, I obtained an image that is quite different as yours: http://i.imgur.com/FNolARd.png I'm trying to achieve a foreground(tubes) vs. background segmentation. – Tin Jun 08 '16 at 12:53
  • That wasn't made with MATLAB but with a personal implementation. I did crop the image by a few pixels as the edges do not continue to the border. –  Jun 08 '16 at 12:56