Graph based segmentation

Graph based segmentation

by Cyril Golaz -
Number of replies: 2

Hello,

In mock exam 2021, question 10, it is said that Graph theoretics methods require some examples of foreground/background pixels to segment. I don't really understand why and from where this could be deduced from the segmentation slides.

Moreover, is it for this reason that in mock 2020, question 3 the correct answer is graph-cut ?

Thanks in advance

Best

Cyril

In reply to Cyril Golaz

Re: Graph based segmentation

by Anar Sabuxi Oglu Rzayev -
For problem 10, refer to the 64th page of the "Regions.pdf" You can see that when doing interactive ST Min-Cut, the user has to manually specify background and foreground pixels as the initialization seeds for the segmentation task. The red and blue pixels are defined in a way to allow the S-T graph cut algorithm to segment the image between foreground (red) and background (blue).

For your second question, yes- that's exactly why the answer is graph-cut. Hope this helps you out!
In reply to Cyril Golaz

Re: Graph based segmentation

by Corentin Dumery -
Agree with Anar, and to add some intuition, both graph-cut and region growing do not have a notion of foreground/background. You need to tell them what you want to be clustered together. So you're saying, "those blue pixels should be together, those red should also be together, now what's the cut in the graph that has the minimum cost?"