Comparative study of texture analysis techniques

Abstract

Texture analysis systems are studied, explaining the various parts which make up a texture analysis system. Also, an evaluation of the ability of four main texture analysis algorithms to perform automatic texture discrimination will be described. The algorithms which will be examined are a Human Vision Based Model, the Localised Fourier Transform method, the Spatial Grey-Level Distribution Method and the Grey-Level Difference method. The localised Fourier Transform method is examined further by considering different windows and their performances. Comparisons based on relative storage requirements, relative speed, relative ability to separate visually distinct textures, theoretical relative performances and finally the ability to correctly classify real textures (paper textures) will be presented. It was found that the Localised Fourier Transform generally performed better, especially when windows having a main lobe/side lobe response close to the Rectangular window are used.