A STUDY ON IMAGE SEGMENTATION USING MOMENTS

Abstract

Image segmentation is a process that searches homogeneity in images and thus can be looked upon as a clustering problem. The moment as an invariant feature has usually been used for object recognition as a descriptor. In this paper, moment is used as a feature for clustering purpose. A Set of moments, as a feature, for each pixel is computed to get homogeneity information of the image. This multidimensional feature vector of each pixel is then classified using non-parametric k-Means clustering method. The noise tolerance capability of the proposed method is checked in comparison to pixel based classification. The window size in computation of moment is a crucial parameter of the proposed method. The comparative study of moments with different window sizes is presented.