Synthetic aperture radar image segmentation based on Markov random field with niche genetic algorithm

Markov Random Field (MRF) has been found to be effective in the domains of image segmentations, since the problems can be simplified to search the optimal label fields. The challenges in MRF image segmentations arise due to the complexity of optimization. Although Genetic Algorithm (GA) has been applied into the image segmentation with MRF, yet most of algorithms defined an individual as a pixel with gray-scales coding, which is powerless to restrain the noise. On the other hand, GA emphasizes the evolution of whole label field, which could cause the over-propagation in some local areas and the convergence to partial optima. To avoid trapping into the local optima, Niche Genetic Algorithm (NGA) is introduced into the MRF image segmentation in this paper. NGA uses the sharing function to restrain the mutation between two individuals with high similarity, which could preserve the diversity of populations. Furthermore, a mechanism of fitness interaction in neighborhoods is proposed to contribute to eliminate the isolated sparkle noise in Synthetic Aperture Radar (SAR) image. The followed segmentation experiment for SAR image proved that MRF segmentation with NGA could reach a satisfied result among the noise restraint, edges preservation and computation complexity.