Image Segmentation Using Markov Random Field Model Learning Feature and Parallel Hybrid Algorithm

In this paper, a new notion of image segmentation using Markov random field (MRF) model learning feature is addressed. The segmentation problem is formulated as pixel labeling problem in a supervised framework. MRF model is employed to model the class labels. This model learns a given training image derived from a class of images. The model having learnt is validated for other images of the same class. The learning problem is formulated using conditional pseudo likelihood (CPL) approach and the parameters are estimated using homotopy continuation method. This learning attribute is exploited to obtain the MAP estimates of the class labels using proposed hybrid tabu search (HTS) and parallel hybrid tabu search (PHTS) algorithms with a view to reduce the computational burden. The performance of these algorithms is compared with that of the simulated annealing (SA) algorithm. This learning feature avoids the parameter estimation of each and every individual image of a class of images.