Two-Stage Texture Segmentation Using Complementary Features

In this paper, a two-stage texture segmentation approach is proposed where an initial segmentation map is obtained through unsupervised clustering of MRSAR features and is followed by self-supervised or bootstrapped classi cation of wavelet features. The selfsupervised stage is based on a segmentation con dence map, where the regions of \high con dence" and \low con dence" are identi ed on the MRSAR segmentation result using multilevel morphological erosion. The second-stage wavelet classi er is trained from the \highcon dence" samples and is used to reclassify only the \low-con dence" pixels. The nal reclassi cation is based on rules that combine minimum distance and spatial constraints. Additionally, an improved coeÆcient feature normalization procedure is used during the classi cation process of both stages. The proposed twostage approach leverages on the advantages of both MRSAR and wavelet features, and incorporates an adaptive neighborhood-based spatial constraint. Experimental results show that the misclassi cation error can be signi cantly reduced compared to morphological cleaning operations alone.