A new texture descriptor based on hexagonal local binary pattern for content-based image retrieval

Abstract

Texture features play a vital role in content-based image retrieval (CBIR) applications. Most texture extraction methods have a low accuracy and high feature vector length. This paper presents a novel hexagonal local binary pattern (HLBP) to extract more informative and compact features from images. To have robust patterns against rotation, rotation invariant hexagonal patterns are presented using cyclic set theory. Texture feature vector is extracted from hexagonal images based on proposed patterns and used in CBIR application. To evaluate proposed method, experiments are performed in five datasets Corel-1k, Brodatz, VisTex, Corel-10k, and STex. The proposed HLBP method outperforms square local binary pattern (SLBP) in images with noise in the terms of precision. The feature vector length of the proposed method is 64, which is much shorter than those in competitive methods and leads to high speed in retrieval phase. The best performance of the proposed method is revealed in texture datasets which achieved the highest precision among all competitive methods.