Leaf analysis for plant recognition

Abstract

Plants are essential resources for nature and people's lives. Plant recognition provides valuable information for plant research and development, and has great impact on environmental protection and exploration. This paper presents a leaf analysis system for plant identification, which consists of three main components. First, given a leaf image, a preprocessing step is conducted for noise reduction. Second, the feature extraction component identifies representative features and computes scale invariant feature descriptors. Third, the matching plant species are identified and returned using a weighted K-nearest neighbor search algorithm. The system is implemented as a Windows phone app and is tested on the LeafSnapdataset[8], an electronic field guide developed by Columbia University and University of Maryland with different combinations of species at various orientations, scales and levels of brightness. The experimental results demonstrate the effectiveness of our proposed framework in plant recognition.