Plant Species Recognition using Local Binary and Local Directional Patterns

Automatic plant species recognition is an intriguing and challenging problem which has vital applications in important real-world areas such as determining plant health, identifying new species and conserving biodiversity. Therefore, deriving an effective plant species representation from images is vital for successful species recognition. In this article the robustness of computer vision texture-based recognition techniques, such as Local Binary and Local Directional Pattern implementations, are evaluated for capturing discriminant textural features of leaf images. Furthermore, classification is done using both K-Nearest Neighbour and Support Vector Machine classifiers on the Swedish and Flavia leaf datasets. The results show that both these local patterns are successful at encoding dominant textural characteristics of leaf images to uniquely identify different plant species. The highest result obtained with the proposed methodology on the Flavia and Swedish leaf datasets are 96.94 % and 98.22 % identification accuracy respectively.