A distinct and compact texture descriptor

• A powerful texture descriptor is developed for texture classification.
• The descriptor is built via fractal analysis on the local binary patterns.
• The descriptor enjoys both high discriminative power and robustness.
• The descriptor is compact and computationally efficient.
• The descriptor demonstrated excellent performance on four datasets.

In this paper, a statistical approach to static texture description is developed, which combines a local pattern coding strategy with a robust global descriptor to achieve highly discriminative power, invariance to photometric transformation and strong robustness against geometric changes. Built upon the local binary patterns that are encoded at multiple scales, a statistical descriptor, called pattern fractal spectrum, characterizes the self-similar behavior of the local pattern distributions by calculating fractal dimension on each type of pattern. Compared with other fractal-based approaches, the proposed descriptor is compact, highly distinctive and computationally efficient. We applied the descriptor to texture classification. Our method has demonstrated excellent performance in comparison with state-of-the-art approaches on four challenging benchmark datasets.