Feature Descriptors based on Circular Forms of Local Patterns for Texture Classification

Abstract

Five novel Local Binary Pattern (LBP) based descriptors are presented in this paper for performing texture analysis of an image. The performance of any texture classification method depends upon its dimensionality. As existing lo-cal based extractor methods generate descriptors with huge dimensions, the classifiers using them will suffer. In order to deal with this problem Uniform Weighted LBP (UWLBP), Strong and Uniform Weighted -LBP (SUWLBP), Uni-form Circular and Elliptical Weighted Texture Matrix (UCEWTM), Strong and Uniform CEWTM (SUCEWTM) and Strong and Uniform Twofold Spatial Weighted Complex Patterns (SUTSWCP) are proposed in this paper. The out-puts of these descriptors are fed into machine learning algorithms like NavieBayes (NB), Multilayer-perceptron (MLP), Ibk and J48. Image datasets like Brodtaz, UIUC, Outex-TC-12, KTH-TIPS, ALOT were used to train and test the models. The five descriptors were combined with MLP and compared. The CEWSTM coupled with MLP attained has attained a classification rate of 93.11 % which is the best out of the proposed five descriptors.