Segmentation and Extraction of Individual Leaves from Plant Images for Species Classification

Plant species classification through the examination of images of plant leaves requires as input an image of a single leaf with no stems or other non-leaf objects. Images of plants, however, usually include more than one leaf, stems, branches, flowers, and other non-leaf objects. For such images each individual leaf needs to be extracted into a unique sub-image, and these sub-images must be cleaned to remove all non-leaf objects. A target leaf could then be selected from the group of sub-images to be provided as the input to the plant species classification program. As a part of the research on this thesis, an algorithm was developed to automate the tasks of detecting and extracting leaf sub-images from plant images and to clean the leaf sub-images by removing all non-leaf objects. To implement the algorithm, software was developed in Java. The proposed algorithm produced at least one perfect leaf result in 18 of the 21 (86%) plant images used in this research, while the remaining three (14%) plant images produced acceptable leaves.