Efficient Algorithms for Image Segmentation

Cooperation with:
Deutsches Herzzentrum Berlin

Project description:

Image segmentation is the first step in vision systems. It consists in extracting important features of images. There are two techniques most commonly used:

The second approach seems to be dual to the first one, but presents some advantages such as producing closed contours (the closure of contours is a difficult problem). But, as no filtering is done in order to eliminate noise, it seems to be less efficient in detecting contours.

We develop efficient algorithms based on solving special instances of the Union-Find problem. They use segmentation and edge detection simultaneously, and thus combine the advantages of both methods and avoid their disadvantages.

Efficiency here not only means theoretical bounds on the complexity but also competitive implementations of the algorithms, first for 2D images and recently also for 3D ones.

figure figure
A 2-D tomography and a segmentation
The bones in a 3-D computer tomography