A Visual Word-based Leaf Classification Scheme

Abstract

Due to the enormous number of plant species and the difficulty of training skilled plant taxonomists, automatic plant classification has recently attracted much attention. Among all the plant organs, leaves are the most widely studied part. In this paper, we propose a leaf image classification method based on visual word. Being different from existing works which use SIFT as local feature descriptor, we use Non-Maximal Suppression to obtain evenly distributed feature points, so that the whole leaf can be well covered. For the detected feature points, we use the MROGH as local feature descriptor and then a visual word-based classification is performed. When constructing global feature for each image, to achieve more reasonable spatial information integrating, we extended the weighting scheme of Spatial Pyramid Matching. Experimental results illustrate the efficacy of our method.