MATLAB Answers

Segmenting the nerve like structures in a low resolution image

4 views (last 30 days)
zohreh hosseinaee
zohreh hosseinaee on 24 Sep 2018
Edited: zohreh hosseinaee on 16 Aug 2019
I have an image like the one attached,
It has low quality and the edges are not very clear. Any idea how to segment the nerve-like structures? I've tried the coherence filter, Frangi filter, and Gabor filter and still, I don't have a satisfying result.
Any help is appreciated.

  0 Comments

Sign in to comment.

Answers (1)

Image Analyst
Image Analyst on 25 Sep 2018
Well you could try anisotropic diffusion (demo attached) or a tophat filter or a Kuwahara filter (demo attached), though I don't hold out great hope for those either. But they're worth a try.
Then there are more sophisticated noise reduction routines like BM3D, non-local means, non-local means 2, UINTA, K-LLD, K-SVD, etc. See the bible of image denoising overviews: https://users.soe.ucsc.edu/~milanfar/publications/journal/ModernTour.pdf by one of the leaders in the field.

  3 Comments

zohreh hosseinaee
zohreh hosseinaee on 25 Sep 2018
Thank you very much, I have tried several denoising and coherence filters and still, I don't have the results I'm looking for. I tried to remove the BG after despeckling and attached are my images. The final image is after despeckling, BG removal and coherence filter. and the diffuse image is the image after diffusion. I'm looking for a method for just segmenting the nerves (The line structures). The is supposed to be automatic so I don't want to do any manual segmentation. I'd really appreciate your help.
Image Analyst
Image Analyst on 26 Sep 2018
I suggest you look at the papers, or hire someone to develop an algorithm for you. This is not simple and would take way too much time - way more than I can donate to you. Plus, it seems like there are judgment calls involved as to whether a little, disconnected bright blob should be considered as part of a nearby nerve trace or not. Good luck though.

Sign in to comment.

Sign in to answer this question.