Automatic Selection of Gabor Filters for Pixel Classification

It is clear that texture plays an important role in both human perception and machine vision tasks and, at the global scale of perception, may be regarded as a carrier of region information, Bovik et al. (1). However, since real-world surface textures are very di cult to model mathematically, no exact de nition exists. Therefore, ad-hoc approaches to the analysis of textures have been used, including; local geometric primitives, Julesz (18); local statistical features, Conners and Harlow (5); random eld models, Geman and Geman (14), Cross and Jain (6) and fractals, Pentland (23).