Gabor filter design for multiple texture segmentation

Abstract

| A method is presented for the design of a sin-gle Gabor lter for the segmentation of multi-textured im-ages. Earlier methods were limited to lters designed for one or two textures or were limited to lters selected from a predetermined lter bank. Our proposed method yields new insight into the design of Gabor lters for segmenting multi-textured images and lays an essential foundation for the design of multiple Gabor lters. In the method, Rician statistics of ltered textures at two dierent Gabor-lter envelope scales are used to eciently generate probability density estimates for each ltered texture over an extensive set of candidate lter parameters. Variable degrees of post-ltering and the accompanying eect on postlter output statistics are also included in the design procedure. The result is a unied framework that analytically relates the texture power spectra, Gabor-lter parameters, postlter-ing eects, and image-segmentation error. Finally, the re-sulting lter design is based on all constituent textures and is not constrained to a limited set of candidate lters.