Local spatial information with bag-of-visual-words model via graph-based representation for texture classification

Abstract

This paper proposes an enhanced feature descriptor for texture classification through graph-based representation. Searching the meaningful texture descriptor is a crucial process in pattern analysis and applications. Graph theory is a model-based approach that applies to texture analysis with outstanding results. Therefore, to develop feature descriptors that are robust against many variations images collected from random viewpoints, change in scale, and illumination remains a challenge for researchers. In this work, we propose an Automatically Local Spatial Pattern Mapping (LSPMAuto ) method based on the spatial-BoVW model that can extract local and global features information from the spatial arrangement of image pixels. The proposed approach is evaluated by using three different texture databases: Brodatz, UIUC, and Outex. The experimental results show that the proposed method can achieve highly discriminant descriptors superior to the other methods. © 2020, ICIC International. All rights reserved.