On the significance of leaf sides in automatic leaf-based plant species identification

Because the front side of a leaf and the underside are functionally very different - the former captures sunlight to produce photosynthesis and the latter absorbs carbon dioxide and releases oxygen and vapor - they typically have different visual features. In this paper we study the significance of leaf sides in visual recognition systems for automatic plant species identification. We measure the accuracy of species identifications with a dataset of 63 species of trees from Costa Rica that includes pictures of both, front sides and undersides of tree leaves. The dataset is used as a global dataset and is also partitioned as two datasets: one of front side pictures and one of underside pictures. Training and testing of different algorithms is performed and their accuracies computed for the group of species and for each individual species. For the tested dataset, leaf side is a significant factor for automatic plant species identification. On the average, and for most cases, underside pictures lead to more accurate identifications.