Classification of Natural Texture Images from Shape Analysis of the Legendre Multifractal Spectrum

Abstract

Classification of images of real textures was performed using the Legendre multifractal spectrum. Novel features have been developed from basic geometrical properties of the spectrum. Consideration was given to practical issues, in particular to the choice of measure. Classification was implemented using discriminant analysis on randomly extracted subimages. A total of 240 cases were considered. Discussion of the results is given along with areas in need of further investigation.