A set of statistical radial binary patterns for tree species identification based on bark images

• Published in: Multimedia tools and applications Vol. 80; no. 15; pp. 22373 - 22404
• Main Authors: Boudra, Safia, Yahiaoui, Itheri, Behloul, Ali
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-06-2021; Springer Nature B.V; Springer Verlag
• Subjects: Artificial Intelligence; Artificial neural networks; Bark; Classification; Comparative studies; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Datasets; Flowers & plants; Identification; Image Processing; Image retrieval; Multimedia; Multimedia Information Systems; Neighborhoods; Neural networks; Representations; Sampling; Signal and Image Processing; Special Purpose and Application-Based Systems; Statistical methods; Texture; Trees; Plant identification; Distribution description; SRBP; Bark texture; ISRBP; Radial Binary Pattern
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-020-08874-x

This paper deals with bark texture representation at high scale-space levels for tree species identification. The proposed approach, named, a set of statistical radial binary patterns (sSRBP), is based on the combination of a novel scale-space sampling and an LBP-like radial encoding in the distribution level. This method aims at capturing and encoding large bark structure information. The multi-scale neighborhood is formed by a set of concentric ring-shaped scale levels, in each of which the intensity distribution is represented by statistical features that provide a compact and information-preserving representation of the neighborhood. Then, the gradual distribution variation over scale levels is encoded by a macro pattern code. For each bark sample, five statistical descriptors are obtained and contribute to enhancing the texture representativeness and discriminative power. We evaluated the performances of the proposed approach on four different bark datasets and found that the novel scale-space sampling significantly improves the bark structure representation leading to enhanced performances and outperforming competitive state-of-the-art LBP-like methods. Furthermore, experiments on the color representation of bark samples improve the performances on challenging bark datasets. Moreover, comparative study between the handcrafted sSRBP texture descriptor and convolutional neural network features shows interesting generalization results on the very large BarkNet dataset.