Texture Analysis Using GMRF Model for Image Segmentation on Spectral Clustering

Abstract

Spectral clustering algorithms are newly developing technique in recent years. In this paper, we derive a new pairwise affinity function for spectral clustering based on a measure of texture features represented by Gaussian Markov Random Field (GMRF) model. This model is used to capture the statistical properties of the neighborhood at a pixel, and then pairwise affinities represented by it can cluster the pixels into coherent groups. Having obtained a local similarity measured by regions of coherent texture and brightness, we use the normalized cuts to find partitions of the image. Experimental results demonstrate that the proposed method is effective and robust for image segmentation.