Local Binary Pattern (LBP) and its derivations for image classification

Abstract

Local Binary Pattern (LBP) has emerged in the early 1990s as one of the most prominent texture descriptors. Over the past two decades, a large number of different LBP variants have been proposed. The original LBP model has been modified on several levels in different ways. Many variants have been proposing in literature, such as uniform LBP on the level of grouping or elliptical LBP (ELBP) on the level of topology, etc.

This thesis is an analytical and comparative study of a variety of different LBP variants in order to reach experimental results that help us to better understand each variant from the studied variants and to determine the appropriate application for each one in the field of image classification based on the texture property.

At the beginning of the study, we discussed the original LBP descriptor proposed in 1994. We conducted the classification on four known databases: (Outex, DTD, Brodatz, Vistex). In order to achieve more classification results for different comparisons, we used two common types of classifiers (SVM, KNN). After that, the scope of the study was expanded to include 12 different variants of the original LBP. This thesis has been backed-up with a deep statistical analysis of the obtained results.