Bark Identification Using Improved Statistical Radial Binary Patterns

In this paper, we explore the texture representation at high scale levels for the application of tree bark identification. We mainly propose an Improved Statistical Radial Binary Pattern (ISRBP) texture descriptor, by introducing a new representation of the scale-space to encode large macrostructures with a low-dimensional representation. The proposed descriptor has advantages of a compact and information-preserving description of large macrostructures, computational simplicity, no preprocessing stage, enhanced texture representativeness and discriminative power. We conducted comprehensive experiments on different bark datasets, and the results show the effectiveness of the new representation of scale-space. In addition, the combination of different statistical radial descriptors provides competitive to high identification rates than state-of-the-art LBP texture descriptors.