Robust Texture Analysis

Abstract

Textures are one of the basic features in visual searching and computer vision. In the research literature, most of the attention has been focussed on the texture features with minimal consideration of the noise models. In this chapter we investigate the problem of texture classification from a maximum likelihood perspective. We take into account the texture models (e.g., Gabor and wavelet models and texture distribution models such as gray-level differences, Laws’ models, covariance models, and local binary patterns), the noise distribution, and the inter-dependence of the texture features. We use the Brodatz’s texture database [Brodatz, 1966] in two experiments. Firstly, we use a subset of nine textures from the database in a texture classification experiment. The goal is to classify correctly random samples extracted from the original textures. In these experiments we use the texture distribution models for extracting features as in the work by Ojala et al. [Ojala et al., 1996] . Secondly, we consider a texture retrieval application where we extract random samples from all the 112 original Brodatz’s textures and the goal is to retrieve samples extracted from the same original texture as the query sample. As texture models we use the wavelet model as in the work by Smith and Chang [Smith and Chang, 1994] and the Gabor texture model as in the work by Ma and Manjunath [Ma and Manjunath, 1996].