Hybrid euler fractional local binary patterns for texture classification

This study introduces Euler Fractional Local Binary Pattern (FGLBP) and its robust variants, FGLBP-M and FGLBP-O, to address the noise sensitivity and detail loss inherent in traditional integer-order LBP. By integrating fractional-order gradients via Euler approximations into the local concave-and-convex microstructure framework, FGLBP captures subtle “micro-topological” features and weak boundaries using a tunable parameter. Experimental results on benchmark datasets demonstrate state-of-the-art performance, achieving 98.26% accuracy on the ALOT dataset and 95.56% on STex. The experiments also confirm exceptional robustness against Gaussian noise, maintaining superior accuracy even under extreme signal-to-noise ratios. Furthermore, the proposed method generates compact feature vectors (1 × 200) and maintains high computational efficiency, consistently outperforming various LBP variants while remaining highly competitive with or superior to sophisticated deep learning architectures such as VGG16 and ResNet101. These findings establish FGLBP as a powerful and practical solution for resource-constrained texture recognition.