Parallel Genetic Algorithm based Textured Image Segmentation using Markov Random Field Model
Pradipta Kumar Nanda, Sucheta Panda, Priyadarshi Kanungo · 2004
In this paper, we address the problem of texture in image segmentation in an unsupervised frame work. Markov Random Field model is employed to model the textured images. The problem is formulated as a pixel labeling problem. The labels as well as the MRF model parameters are assumed to be unknown. A coarse grained notion based Parallel Genetic Algorithm (PGA) is proposed to estimate the pixel label together with the model parameters. With the evolution of the algorithm, the model parameters, starting from an arbitrry value, evolve to converge to the optimal estimates. The algorithm starts with arbitrary pixel labels and evolve to converge eventually to stable labels. In the proposed PGA algorithm the crossover and mutation probabilities are adaptive with the progress in generation. The algorithm is validated for synthetic as well as real images.