Leaf classification using structure features and Support Vector Machines

In Thailand, there are a lot of near extinction herbs, because they are used as an essential resource for food and medicine industry. Before they become extinct, the herb information needs a systematic collection. This paper is proposed a herb leaf classification method for making an automatic categorization. In digitization step, the leaves with white background are photographed with digital camera. As a preprocessing step, apex and leafstalk are removed with a histogram-based method. To describe leaf characteristic, three types of ratio; aspect ratio, slice ratio and radius ratio are measured. These features are scale in variant and -/+15 degrees tolerance of rotation. In experimental results, our technique performs 95 percent accuracy on 10 classes of leaf type.