Energy based wavelet and multilevel classifier for efficient leaf recognition

Abstract

Computer Aided Leaf Recognition (CALR) is an important field of research that provides tools for forestry, agriculture and pharmacy. Due to the deterioration of environment, rare species of plants are at the brim of extinction. Investigation of rare plants though CALR can subsidize to environmental protection. Generally, CALR system consists of four main steps, such as, enhancement, segmentation, leaf feature extraction and classification. Preprocessing step enhances the leaf image by removing noise, modifying contrast and highlighting boundaries. To separate leaf image from the background, the CALR system uses clustering combined with Energy Based Wavelet (EBW) segmentation. Optimized Principal Component Analysis (OPCA) is used to extracts 28 features falling under five categories, namely, geometry, color, texture, fractal and leaf specialization. A two-level classifier is used to improve the accuracy of recognition process. A refined training set is generated during the first level, and it is used to train the second level classifier. Standard leaf image dataset and real leaf image dataset are used to evaluate the performance of proposed algorithm. This leaf recognition model is effective in discriminating leaves and identifying plant. Hence, taxonomists can use this system to identify precious plant leaves in order to save them.