A novel image segmentation algorithm based on Hidden Markov Random Field model and Finite Mixture Model parameter estimation

Abstract

Hidden Markov Random Field (HMRF) model and Finite Mixture Model (FMM) parameter estimation algorithm provides an interesting framework for image segmentation task, hence a technique that capitalizes on the benefits of both algorithms would achieve better performance. In this regard, we propose a new segmentation algorithm which combines with HMRF model and FMM parameter estimation algorithm. Firstly, we use a real-coded genetic algorithm based FMM to estimate image parameters. Secondly, according to the estimated image parameters, image pixels are classified into different classes through the HMRF segmentation framework. The performance of the proposed algorithm is tested on Berkeley image segmentation dataset. Experimental results have confirmed that the proposed algorithm offers a useful improvement of the segmentation accuracy over competing methodologies.