Leaf-Based Plant Identification Through Morphological Characterization in Digital Images

摘要

The plant species identification is a manual process performed mainly by botanical scientists based on their experience. In order to improve this task, several plant classification processes has been proposed applying pattern recognition. In this work, we propose a method combining three visual attributes of leaves: boundary shape, texture and color. Complex networks and multi-scale fractal dimension techniques were used to characterize the leaf boundary shape, the Haralick's descriptors for texture were extracted, and color moments were calculated. Experiments were performed on the ImageCLEF 2012 train dataset, scan pictures only. We reached up to 90.41% of accuracy regarding the leaf-based plant identification problem for 115 species.