Phytomorphological graph construction for leaf identification of a 2D monocotyledon image

We propose a graph construction method for automatic leaf identification of a monocotyledon image. Leaf identification is one of key technologies to acquire plant phenotypes such as a leaf length, a leaf count, and a growth rate, because it is important in the field of high-throughput phenotyping to repeatedly analyze the structure of plants and its phenotypes from a huge number of crops. However, it is challenging to identify individual leaves from a monocotyledonous plant image due to their complicated occlusion and similar colors. So we choose a graph structure as a technique to overcome the leaf occlusion and the color similarity between leaves. In order to construct a graph from a raw input image of monocotyledonous plants such as rice plants, we apply a modified GrabCut algorithm to extract a plant region considering morphological and color characteristics of plants, then compute a skeleton from the extracted plant region, and finally construct a graph from the plant skeleton using a skeleton following algorithm and the concept of neighbor group, which is called a phytomorphological graph. Experiments show that our proposed method effectively constructs a topological graph which reflects the architecture of a plant from a single 2-dimensional image, and facilitates automatic leaf identification which enables us to take an accurate and efficient high-throughput measurement of phenotypes for plants.