Gradient ternary transition based cross diagonal texture matrix for texture classification

This paper derives a new local descriptor gradient ternary transition based cross diagonal texture matrix (GTCDTM) for texture classification. This paper initially divides the image into a 3x3 window in an overlapped manner. On each 3x3 window, this paper computes the gradient between center pixel and each sampling point of the window. This paper divides the gradient window into cross and diagonal matrices and computes gradient transition (GT) cross unit (GTCU) and GT diagonal unit (GTDU). The GT’s are derived by computing relationship between adjacent gradient pixels of cross and diagonal matrices in a clock wise manner. This research derived GTCDTM by computing the occurrence frequencies of GTCU vs. GTDU. The gray level co-occurrence matrix (GLCM) features derived on the proposed GTCDTM descriptor derive the feature vector. The proposed descriptor is tested on the popular databases using machine learning classifiers and equated with state of art local based methods. The results indicate the efficacy of the proposed method.