Texture in images: Algorithms for comparison and segmentation

Abstract

The extraction of features that are sensitive to texture in an image has been the subject of intensive investigations in recent years. Recently, several important industrial applications based on the texture of a surface or texture in a scene have been identified. Many of these applications involve classification of texture, comparison of two texture samples or segmentation of an image into texturally homogenous regions. In this paper, the maximum likelihood technique has been adopted to enable comparison of two textures (similarity measure) as well as segmentation of a given image into texturally homogenous regions. In addition, features are derived from the gradient of the image rather than the spatial gray-level co-occurrence matrix. A new measure of similarlity termed, "the similarity index (SI)" has been derived for comparing two textures (e.g. homogeneity of a painted surface). Experimental results with a variety of textures have demonstrated the feasibility of the new approaches taken.