On Computation of Texture Descriptors from Sum and Difference Histograms

Abstract

Texture information is used in some computational activities such as pattern recognition and segmentation. One of the most used methods to obtain this information uses the Haralick descriptors that are calculated from the Gray Level Occurrence Matrix (GLCM) that has a quadratic cost. On the other hand, in this work we discuss an approach that has linear cost and achieves the same results using Sum and Difference Histograms (SDH). A set of nine equations from SDH presenting similar results to GLCM have been introduced some time ago, however, their differences may vary greatly in absolute values. This work presents an evolution in this direction that allows obtaining descriptors numerically equal or with a very small variation around 10−1. In addition, another six new equations are added in this study for other descriptors. All results from SDH use less computational resources and present the same behavior as those obtained by GLCM. This work illustrates the above-mentioned statement.