Semi-supervised segmentation of textured images by using coupled MRF model

Markov Random Field (MRF) is extensively used in model-based segmentation of textured images. In this paper, we propose a coupled MRF model and adopt the MAP-MRF framework to solve the semi-supervised segmentation problem. The observed image and the desired labeling are characterized by the Conditional Markov (CM) model and the Multi-Level Logistic (MLL) model, respectively. The parameters of CM models are estimated as texture features, and contextual dependent constraints are imposed to the object function by the MLL model. Different from existing methods, the two MRF models are mutually dependent in our approach and therefore texture features and the labeling must be optimized simultaneously. To this end, a step-wised optimization scheme is presented to achieve a suboptimal solution. The proposed algorithm is compared with a simple MRF model based method in segmentation of Brodatz texture mosaics. The experimental results demonstrate that the novel approach can differentiate textured images more accurately.