Noise-robust statistical feature distributions for texture analysis

A novel image feature extraction methodology is proposed in this study. By incorporating fuzzy logic into the well-established Local Binary Pattern (LBP) approach we derive statistical feature distributions suitable for noise-robust texture representation. The proposed Fuzzy Local Binary Pattern (FLBP) approach is based on the assumption that a local image neighbourhood may be characterized by more than a single binary pattern. The effectiveness of the proposed methodology is demonstrated by classification experiments on noise degraded Brodatz textures. The classification performance obtained with the FLBP features was higher than the one obtained with the original LBP features for various noise levels.