IR image segmentation using GA-MRF with Neighborhood Labels Coding

Comparing with Simulated Annealing (SA) that is often used in the image segmentations based on Markov Random Field (MRF) models, Genetic Algorithm (GA) has been applied into reducing the computation complexity of optimization. However many scholars used GA as an optimal tool that based on the gray-scale values of pixels as individuals and limited the mutations and crossovers with the gray-level coding, which caused these algorithms sensitive to noise especially to the multiplicative noise. To avoid trapping into the low-grade imitations of canonical GA with gray-level values of pixels, the labels coding of individuals in a neighborhood instead of the gray-scale values coding is proposed in this paper. And the mutations and crossovers with labels coding in a neighborhood increased the efficiency of searching optimal and preserved the original information of images. The followed experiments with Infrared (IR) image segmentations proved that the proposal algorithm approached an acceptable result among the noise restraint, edges preservation and computation complexity.