Edge-Based Segmentation of Textured Images Using Optimally Selected Gabor Filters

In this paper, we propose a technique for segmenting visual textures using features extracted from the reponses of Ga,bor filters, appropria.tely selec tecl to be tuned to texture components of the input image. In the proposed method segmentation is achieved by detecting boundaries between adjacent textured regions, and this segmentation algorithm works as follows. The input image is first filtered using a small set of Gabor filters, each tuned to one of the textures composing the original image. Abrupt changes in the obtained Gabor filter output images are found by detecting the underlying local extremas. For this purpose, a gradient operator is applied to output ima.ge of each Gabor filter, yielding a. set of gradient images. The textcrre gradient is sul>sequently obtained by grouping gradient images from all channels. Thresholding the tex ture gradient and thinning the result yields the expected texture boundaries. Experimental results on synthetic and natural textures, demonstrate the efficacy of the proposed technique.