Recognition of leaves based on morphological features derived from two half-regions

Leaf recognition systems can be used for automatic plant taxonomy and provide understanding and managing of plants in botany, medicine, industry and food sector. Trees and flowery plants can be classified by using leaf recognition. This paper proposes a simple method based on bisection of leaves for recognition. After preprocessing techniques are applied for leaves, 7 low-cost morphological features are extracted which are used in the literature. We produced 3 additional features using half leaf images. Most of leaf species have morphological structure that resembles each other a lot. For these leaves, while structural features of one half resemble, features of other half differ. Taking advantage of this knowledge, leaf is oriented according to its major axis and two parts are acquired by slicing leaf on its centroid vertically. Area, extent and eccentricity features are extracted for each part and their proportions to each other are taken as new features in this study. These all 10 features are used as an input to probabilistic neural network (PNN). PNN is trained with 1120 leaf images from 32 different plant species which are taken from FLAVIA dataset. 160 leaf images from the plant species are used for testing. Our experiments and comparisons show that method based on half leaf features has reached one of the best results in the literature for PNN with 92.5% recognition accuracy.