Branching points(OpenCV,C++)

Question

I want to identify the branching points of the lightning in this image:

https://i.stack.imgur.com/PXujf.jpg

What I did first was threshold the image such that I get the lighning part of the image and discard the backround. This is the result

https://i.stack.imgur.com/IYNTi.jpg

I used the threshold function in openCV and the resulting image is pretty much bad as the quality is lost, the branches are no longer visible.

Ok, basically I have 2 problems:

How can i properly segment the image such that the lightning part of the image is properly captured.
How can I then, identify the branching points? For every branch point i want to draw a red circle over it.

Thanking you in advance

`cvAdaptiveThreshold` should work better than normal threshold in this case. — Piotr Praszmo, Sep 22 '11 at 12:29
Thanks Banthar, but how is it different to the normal thresholding? — user854576, Sep 22 '11 at 12:39
http://i.imgur.com/lTRC5.jpg It will better separate lightnings from sky, jpeg artifacts will be more visible though. — Piotr Praszmo, Sep 22 '11 at 13:10
Superb Banthar, step into the right direction, thumbs up bigtime (",) — user854576, Sep 22 '11 at 13:18
Now if I perform morphological operation on the picture, like erode it, will it remove the small black dots in the image? — user854576, Sep 22 '11 at 13:24
It should. You can also try to blur it before thresholding or tune threshold parameters. — Piotr Praszmo, Sep 22 '11 at 13:29

score 0 · Answer 1 · answered Jan 24 '13 at 11:59

Segmentation / thresholding:
I would give this a try. They also had a NIPS2012 workshop (DISCML) paper on image segmentation that seems to handle quite elegantly thin elongated objects (like the lightnings in the picture).

Branching points:
Once you have a good mask you can use morphological operations to extract the branching points (Matlab code):

 bw = myGoodSegmentation( img ); % replace with whatever segmentation/thresholding that works best for you.
 tbw = bwmorph( bw, 'skel', inf );
 [y x] = find( bwmorph( tbw, 'branchpoints' ) ); % get x,y coordinates of branch points

Branching points(OpenCV,C++)

1 Answers1

Linked