Directional analysis of image textures for feature extraction and segmentation

This paper describes a new approach for texture feature extraction, appropriate for unsupervised image segmentation applications. The pixel features are obtained by using directional filters to analyze the given image to a set of subimages, each one containing an isolated angular section of the initial image spectrum, and estimating the pixel local energy in them. By incorporating the proposed feature extraction technique into a single multicomponent texture segmentation procedure, some experiments with texture separation have been implemented and the effectiveness of the presented method has been tested. The results are presented and a comparison with feature extraction techniques, based on the discrete wavelet decomposition of the image, is made.