Inner & Outre Chord Matrix: One Efficient Shape Desiptor Used Plant Leaf Identification

Description and match of the shape in the pattern recognition and the image processing is a very important research subject, which has been widely used in various fields of medicine, biology, agriculture, industry, etc. Plant classification is one of the very valuable application. Main work done includes in this thesis:(1) This thesis analyzes various contour shape description methods, in particular multiscale distance matrix. This thesis analyzes its limitation when applied to identify plant leaves, and points out that the descriptor is not able to describe the concave-convex characteristic of profile shape effectively.(2) This thesis proposes one new shape descriptor-Inner&Outer Chord Matrix. IOCM can effectively extract the concave-convex characteristic of profile shape, while human perception system is very sensitive to this characteristic. In view of the cognition, the ability of IOCM identifying the shape will be greatly enhanced. The method is applied to Swedish and Leaf 100 plant leaf image library, comparing with similar MDM method and classical Fourier Descriptor, IOCM achieves better recognition rate. To compress space, the classical Linear Discriminant Analysis and Maximum Margin Criterion method which is proposed recently feature reduction methods are applied in this paper. Experimental results show that after the characteristic dimension reduction, IOCM still maintains good discernment.