Color Images Co-segmentation Based on Fuzzy Local-Entropy Classification

In this paper, we are interested in the unsupervised co-segmentation of color images. Solving this co-segmentation problem returns usually to optimize an energy function, which evaluates the similarity between the similar foreground objects in the input images. The objective is to evaluate the correspondence of foreground objects that penalizes the dissimilarity between them. To assess this correspondence existing techniques simply compare the histograms in the absence of any information of spatial coherence. The purpose of this paper is to integrate spatial information in order to avoid false detection. Indeed, in addition to the integration of the spatial information thanks to the use of the local entropy during the histogram computing, the main contribution of the proposed technique resides in the fuzzy local-entropy classification which allows to model the ambiguity of a pixel membership to a histogram bin. In particular, this permits to minimize over-segmentation and noise effects on the final co-segmentation results. Recorded results and the comparative study prove the accuracy of the proposed technique for color images co-segmentation.