Texture classification is an important task in computer vision and image analysis with applications in industrial inspection, medical imaging, and remote sensing. Local Binary Pattern (LBP) descriptors are widely used due to their computational simplicity and strong discriminative capability; however, they are sensitive to image noise and scale variations. This paper proposes a Noise-Robust Scale-Selective Local Binary Pattern (NR-SSLBP) with Multi-Sample Feature Averaging to improve robustness in texture classification. The method incorporates noise-reduction preprocessing to stabilize local intensity comparisons and employs multi-scale analysis to capture texture patterns at different spatial resolutions. Dominant patterns across scales are selected to construct discriminative features, while multi-sample feature averaging reduces instability caused by noise and small texture variations. Experimental results show that the proposed NR-SSLBP descriptor improves classification accuracy and robustness while maintaining the computational efficiency of LBP-based methods.
Rao et al. (Tue,) studied this question.