Leaf Identification using Harris Corner Detection, SURF Feature and FLANN Matcher

Abstract

Leaf Recognition is very important in agriculture for identification of plants. Leaves of various plants have unique characteristics which are to be used for categorization. Out of the different features, leaf vein is one of the prominent biometric feature. Extracting leaf vein and perform classification based on these features leads to more accurate identification of plants. In practice, due to change in various lighting conditions and orientations, the extraction of leaf vein becomes difficult. This work focuses on extracting veins using ridge orientation and frequency estimation using region mask which brings out good quality vein structure under different conditions. The vein structure thus obtained is used for identifying keypoints using Harris corner detector. Features are extracted from the keypoints using SURF feature extraction method and finally the trained and query images are compared to identify the correct leaf species using FLANN matcher. Flavia leaf image database with 32 different species are used and an accuracy of 98.75% was resulted. The proposed methodology can be used for plant leaf identification in real world for identifying medicinal plants and other category of plants. This method can be used for identifying veins of dry leaves which can further extract the features and identify the species.