Genetic algorithm based segmentation of high resolution multispectral images using GMRF model

This paper examines hybrid genetic algorithm and Gaussian Markov random field model based method for unsupervised segmentation of multi-spectral textured images. It also evaluates the popular unsupervised image segmentation approaches, Genetic algorithm (GA) based clustering and simple Gaussian Markov random field (GMRF) model with the hybrid GA-GMRF method for high spatial resolution textured imagery. Each method is described and the compatibility of each method with the textured image is examined. It is observed that GA based clustering is more suitable for medium resolution imagery and for images without textures. GMRF model using iterated conditional modes (ICM) algorithm which gives desirable results for textured images, requires several iteration steps to approximate near optimal solutions. The hybrid GA-MRF method, in which the powerful global exploration of GA is used to initialize the ICM algorithm, has found more promising and gives improved results in terms of both accuracy and time complexity than the two other methods for multi-spectral textured images.