The main subject of this work is a theoretical and experimental comparison of two popular image segmentation algorithms: fuzzy connectedness FC and graph cut GC. On the theoretical side, our emphasis is on describing a common framework in which both of these methods can be naturally expressed. We give a full analysis of the framework and describe precisely a place which each of the two methods occupies in it. Within the same framework, other region based segmentation methods, e.g., watershed and some versions of level set algorithms can also be expressed. A special emphasis is given to delineation algorithms (i.e., segmentation algorithms returning only one object), since this makes the comparison clearer and since GC works well only in such a set-up. The main distinguishing characteristics of these methods are also analyzed theoretically.
An experimental comparison of the performance of FC and GC algorithms is also included. This concentrates on comparing the running times of actual (as opposed to provable worst scenario) algorithms, as well as on the influence of the output depending on the choice of seed points specified. Some distinguishing characteristics of these algorithms analyzed theoretically are verified and illustrated empirically in the experiments.
Conference Proceeding reprint.
Last modified March 10, 2011.