A fuzzy logic model based Markov random field for medical image segmentation

Abstract

Fuzzy logic incorporates human knowledge into the system via facts and rules and hence widely used in image segmentation. Another successful approach in image segmentation is the use of Markov random field (MRF) to incorporate local spatial information between neighboring pixels of an image. This paper studies the benefits of these two approaches and combines fuzzy logic with MRF to develop a new adaptive fuzzy inference system. The premise part of each fuzzy if–then rule in our approach adopts MRF to utilize the spatial constraint in an image, while the consequent part specifies the pixel distance map. Unlike other fuzzy logic models that require training data that is not always available to train the system before segmenting an image, we propose an unsupervised learning algorithm for automatic image segmentation. To impose the spatial information on the fuzzy if–then rule base, a new clique potential MRF function is proposed in this paper. Our approach is used to segment many medical images from simulated brain images to real brain ones with excellent results.