Tree recognition based on leaf images

This paper addresses an image-based tree recognition method using shape features and color features of leaf images. Our method requires two leaf images including a front leaf image and a rear leaf image which are placed on a white background. Firstly, a leaf region is automatically extracted using a graph cuts-based method. Next, sixteen shape features, six color features, and one size feature are calculated from the extracted leaf region. These features are fed to random forests. In our experiments, 96 classes (92 species) with 20 samples are collected in fields around our campus. The experimental results reveal that the recognition accuracy of the rear leaf is higher than that of the front leaf.