An approach to texture-based image recognition by deconstructing multispectral co-occurrence matrices using Tchebichef orthogonal polynomials

The existing use of summary statistics from co-occurrence matrices of images for texture recognition and classification has inadequacies when dealing with non-uniform and colored texture such as traditional `Batik¿ and `Songket¿ cloth motifs. This study uses the Tchebichef orthogonal polynomial as a way to preserve the shape information of cooccurrence matrices generated using the RGB multispectral method; allowing prominent features and shapes of the matrices to be preserved while discarding extraneous information. The decomposition of the six multispectral co-occurrence matrices yields a set of moment coefficients which can be used to quantify the difference between textures. The proposed method, when tested with a subset of the Vision Texture (VisTex) database and a collection of `Batik¿ and `Songket¿ motifs, yielded promising results of 99.5% and 95.28% classification rates respectively using the 3-nearest neighbor classifier.