ANALYSIS AND COMPARISON OF SPACE SPATIAL-FREQUENCY AND MULTISCALE METHODS FOR TEXTURE SEGMENTATION

We investigate the use of space/spatial-frequency and multiscale analysis methods for texture segmentation, with emphasis on the two-dimensional Wigner-Ville distribution and Morlet wavelet transform. For these two methods, the discrete versions that are necessary for numerical implementations are discussed. Texture segmentation paradigms making use of local spectral measurements from these two types of representations are described. The interest of the proposed spatial-frequency- and scale-based segmentation methods is illustrated with the aid of examples on both synthesized and natural images, and their segmentation performance is analyzed and compared.