Abstract
Leaf morphological characters are a useful visual guide for constructing relationships between different plants and between plants and their environment. However, extracting and analysing these characters are carried out manually by botanists, which is a painstaking and time-consuming task. One way to accelerate and broaden the use of these characters is to automatically extract them directly from images. An indispensable step toward this goal is to automatically detect leaf parts (petiole, blade, base, apex, rachis) since foliar characters are key descriptions about their shapes. In this paper we present a novel approach that addresses this problem. It is based on two types of symmetry: the first is local translational symmetry (for petiole, rachis detection). The second is local symmetry of depth indentations (for base and apex detection). The main advantage of this method is its accuracy and its robustness to shape variability. This is confirmed by the high rate of correct detections (more than 90%) obtained for a large number of leaf species.