Texture segmentation by clustering the phase of HOS cepstra

Many applications require identification, segmentation and deconvolution of textures and detection of the objects of combined patterns. Frequency-based analysis of patterns (see Tan, T.N. and Constantinides, A.G., 1990) does not avoid the redundancy which highly deteriorates the process. On the other hand, spatial quantifiers (see Sanei, S. et al., 1991) rely on the rough estimation of the model parameters, which are not robust enough to details and alteration of the regions size. By using HOS (see Nikias, C.L. and Petropulu, A.P., 1993), classification based on variation in minimum and maximum phase of cepstra, through a spatial-based sliding window incorporates both space and frequency differences. A small number of minimum and maximum phase coefficients are then evaluated for a sliding window of fixed size. The results show an attractive implementation of HOS estimation in texture segmentation