P-type based dimensionality reduction for open contours of Colombian Páramo plant species

Contour analysis for interpretable leaf shape category discovery

“…The proposed method can test existing manuals and validate whether they contain the information necessary to classify under-studied leaf morphology or propose a classification form that follows a rigorous mathematical method to avoid ambiguity when choosing a category. To discover these categories, we used a highly flexible low-level representation space that captures biologically meaningful information of the leaf border, in particular, its large and fine variations [25]. The proposed representation captured a broad set of lamina border variations exhaustively in the Fourier harmonics, providing a rich morphospace to represent possibly unknown sample morphologies.…”

“…For instance, the approach can be useful when the contours present different lobular compositions, or in sessile leaves, which do not have petiole resulting in open contours. This kind of description is essential also for the description of the external morphology on the leaf sheet [25]. A three dimensional space obtained by PCA embedded the contour representation and a non-supervised clustering algorithm was used on this representation space to infer the corresponding shape categories.…”