Abstract: In order to recognize plant species from their shapes, countours of their leaves can be used. Leaves can be classified using their structural property that a leaf consists of triangular pieces that protrude around a polygon. However their shapes have variations even though they belong to the same species. Variations of leaf contours can be represented statistically. However there is variation caused by the large deformation of structural elements of leaves. Because tips of leaf veins can bend according to their environment, shapes of pieces can be deformed largely and influence the descriptions of leaves. It is desirable such deformed shapes be modified to the original undeformed shapes of the leaves in oder to reduce the variation of the leaf shapes.
A leaf is considered to be composed of a number of overlapping leaflets wich have a single apex and have symmetric shapes with respect to their veins basically. In this paper, a method that normalizes shapes of leaves is presented using the symmetry of each leaflet with respect to its vein. Recognition using normalized shapes of leaves shows improved results compared with the method using unnormalized leaves.