Optimal Gabor filter design for texture segmentation using stochastic optimization

In this paper. we consider the issue of designing a single Gabor filter for multiple texture segmentation using a systematic optimization algorithm. T he proposed algorithm is a stochastic search technique based on the simulat ed annealing (SA) procedure. It embeds the pattern search into the SA proce dure as the move generation mechanism to accelerate the search. The selecti on objective for a best Gabor filter is based on the Maxmin principle that maximizes the minimum energy ratio of any two distinct textures in question . The objective makes the energy responses between different texture classe s well separated. Therefore, a simple thresholding scheme can be directly a pplied to partition an input image into differently textured regions. The e xperiments on a number of bipartite, tripartite and quadripartite textured images have shown promising results using the proposed method. (C) 2001 Els evier Science B.V. All rights reserved.