Parameter Selection in Elliptical Fourier Series for Leaf Classification

Abstract

Elliptical Fourier series method is a relatively old but powerful method in contour analysis. We have previously implemented this method as a module for a computer program called “BioMorph”. There are two significant parameters in this module; the first one is the number of harmonics (sine-cosine terms) that are calculated to form a function which represents the contour. The second parameter is the number of samples that are collected by sampling the function which is a continuous function of sums of sine and cosine terms. In this study, the effects of those two parameters to leaf classification are investigated. Also classification results obtained using reconstructed coordinates and using magnitudes of Fourier coefficients are compared. While the highest classification accuracy is obtained as 75,197% with minimum 15 sine-cosine terms and 20 samples per contour using reconstructed coordinates, it is obtained as 69,953% using magnitudes of Fourier coefficients with minimum 29 terms.