A Novel Approach to Object/Background Segmentation Based on the Probabilistic Graphical Model
Qiuxu Li, Jieyu Zhao · 2009
Graph cut as a powerful optimization technique for minimizing MRF (Markov Random Field) energy functions has been successfully applied to image segmentation. In this paper, we adopt an MRF model for object/background segmentation. The theoretical framework is based on maximum a posterior estimation via the graph-cut energy optimization method. Parameters are estimated with a novel parameter estimation algorithm. The novel parameter estimation algorithm is a variant of the expectation maximization (EM) algorithm with prior influence factors. Characteristic features related to the information in color, texture and position are extracted for each pixel. Experimental results demonstrate the effectiveness of our approach.