Fig. 6From: Contour analysis for interpretable leaf shape category discoveryGraphical representation of the strategy for category discovery in leaves dataset. The leaf contours in the dataset were obtained by using binarization and contour extraction. This shape information was represented by a complex Fourier transform. A set of representative harmonics of the leaf information were selected. Following, a dimensionality reduction method was applied to the selected harmonics. Finally, an adaptive kernel density estimation method was used to determine the shape categoriesBack to article page