Individual leaf identification from horticultural crop images based on the leaf skeleton

Complex scenarios make individual leaf identification challenging, such as messy distributions, fuzzy edges, overlapping, and weeds. In this study, we solved this problem by using the skeleton to estimate the locations and directions of the leaves. We propose a segmentation algorithm based on similar tangential direction (TD) to retrieve skeletons, and we then use the relative moment and the number of pixels for directions to compute the leaf distribution. Experimental results obtained based on 12 types of leaves demonstrated that our proposed system can automatically identify individual leaves. In addition, our system can obtain robust performance with complex horticultural crop images.