Texture segmentation using a diffusion region growing technique

In this paper we consider the segmentation of textured images using a diffusion region growing method. This method is able to grow several separate (but similar) regions simultaneously, and is very well suited to parallel implementation. Recent research, both in computer and human vision, suggests the use of spatial/spatial-frequency (s/sf) image representations as the basis for such tasks. The representation we have used is an approximation to the Wigner distribution, an s/sf representation that has recently been suggested as corresponding to the early and preattentive stage of visual processing. Experimental results on both synthesized and natural textures are shown.