LEAF SHAPE MODELING AND ANALYSIS USING GEOMETRIC DESCRIPTORS DERIVED FROM BEZIER CURVES

Abstract

Algorithms were developed to extract the leaf boundary of selected vegetable seedlings. The leaf boundary wasfitted with Bezier curves, and geometric descriptors of the leaf shape were then derived. In the boundary extraction phase,a color image of a seedling leaf was first obtained and segmented into binary images. The centroid of each leaf was found,and the boundary signature was determined as a function of radial angle. Points at the leaf apex and leaf base were thenlocated from the curvature of boundary points. These points were used as the initial control points of the Bezier curves to fitthe leaf boundary. Given a mathematical representation of the leaf boundary (i.e., the Bezier function) and the coordinatesof the normalized control points, the shape of a leaf may be quantified and reconstructed. Leaf features, including apex angle,base angle, control line ratios, and fitting error, were subsequently derived from the fitted Bezier curves. These features areindependent of size and orientation. The efficacy of using Bezier curves to model leaf boundary and derivation of leaf featureswas examined by comparing the actual leaf area and the modeled leaf area. Experiments were also performed to demonstratethe use of derived leaf features for plant identification. A classification rate of 95.1% was achieved in classifying four varietiesof vegetable seedlings using a backpropagation neural network. The descriptors derived from Bezier curves provide a meansfor leaf shape description that may be useful for plant identification and growth measurement. Leaf shape modeled by simpleBezier curves also contributes a significant data reduction, compared with that using discrete boundary points, whilepreserving reasonable accuracy.