CN-LBP: Complex Networks-based Local Binary Patterns for Texture Classification

TL;DR

This paper introduces CN-LBP, a novel texture descriptor combining complex network features with local binary patterns, which enhances texture classification accuracy and robustness to noise.

Contribution

The paper proposes a new texture descriptor that integrates complex network measurements with LBP, providing more detailed information and improved classification performance.

Findings

Significantly outperforms existing LBP variants in classification accuracy.

Demonstrates robustness to imaging noise and variations.

Effective across multiple datasets.

Abstract

To overcome the limitations of original local binary patterns (LBP), this article proposes a new texture descriptor aided by complex networks (CN) and LBP, named CN-LBP. Specifically, we first abstract a texture image (TI) as directed graphs over different bands with the help of pixel distance, intensity, and gradient (magnitude and angle). Second, several CN-based feature measurements (including clustering coefficient, in-degree centrality, out-degree centrality, and eigenvector centrality) are selected to further decipher the texture features, which generates four feature images that can retain the image information as much as possible. Third, given the original TIs, gradient images (GI), and generated feature images, we can obtain the discriminative representation of texture images based on uniform LBP (ULBP). Finally, the feature vector is obtained by jointly calculating and concatenating the spatial histograms. In contrast to original LBP, the proposed texture descriptor contains more detailed image information, and shows resistance to imaging and noise. Experiment results on four datasets demonstrate that the proposed texture descriptor can significantly improve the classification accuracies compared with the state-of-the-art LBP-based variants and deep learning-based methods.