Leaf recognition and segmentation by using depth image

Measuring the geometric structural traits of plants, especially the shape of leaves, plays an important role in the agricultural science. However, most existing techniques and systems have limited overall performance in accuracy, efficiency and descriptive ability, which is insufficient for the requirements in many real applications. In this study, a new kind of sensing device, the Kinect depth sensor which measures the real distance to objects directly and is able to capture high-resolution depth images, is exploited for the automatic recognition and extraction of leaves. The pixels of the depth image are converted into a set of 3D points and transformed into a standard coordinate system after ground calibration. Leaves are extracted based on the height information and a hierarchical clustering algorithm, which combines the density-based spatial clustering algorithm and the mean-shift algorithm, is proposed for the automatic segmentation of leaves. Experimental result shows the effectiveness of our proposed method.