EAGLE: A novel descriptor for identifying plant species using leaf lamina vascular features

The timely and accurate identification of plant species is a persistent challenge as pressure from human activity threatens global flora biodiversity. Most existing research on computer based plant species identification has focused on using leaf contour, signature and spectral analysis techniques alongside textural properties of the leaf lamina. However, these global feature based methods often suffer from limited discriminative ability and scalability, particularly on mobile devices for use in the field. In this paper, we propose a novel descriptor named EAGLE and employ the popular Bag of Visual Words (BoVW) model to achieve scalable and effective species identification suitable for implementation on mobile devices. EAGLE exploits the vascular structure of a leaf within a spatial context, where the edge patterns among neighbouring regions characterise the overall venation structure and are represented in a histogram of angular relationships. Experimental results on the widely used Swedish leaf image database demonstrated the EAGLE descriptor is able to boost the performance of the effective SURF local descriptor by 6%.