Hierarchical segmentation of digital mammography by agents competition

Abstract

A new hybrid approach for mammography segmentation is suggested in this work. The segmentations proceed by refining successively, in a way that macro regions at each step are recovered. Two segmentation techniques, thresholding and Markov field, enter in competition to segment theses regions. The technique which gives the best results according to a given criterion will take on the process. However, using the same segmentation for all macro regions is not necessary. Thus, the new obtained regions will become candidates for a novel segmentation. The process continues till verification of a stop criterion.