Texture Pattern Classification via Weibull Distribution-based Image Statistical Modeling

Abstract

A novel image texture pattern classification (ITPC) method based on statistical modeling of image spatial structures (ISSs) is presented due to the perceptual perspective that the visual appearance of image texture pattern (ITP) visually depends on the random organization of local homogeneous fragments. Firstly, the oriented Gaussian derivative filters (OGDFs)-based ISS characteristics expression method is introduced. Then, the Weibull distribution (WD) behavior of ISSs, boiled down to the randomly-distributed local homogeneous fragments, is explicated by the sequential fragmentation theory (SFT). Finally, the WD-based ISS feature parameters, directly related to the human vision perception(HVP) system, are extracted and used to ITPC by a partial least squares-discriminant analysis (PLS-DA)-based classifier. The effectiveness of the proposed ISS-based ITPC method is verified by extensive experiments on three different texture image databases and compared with some classic statistics-based ITP identification methods.