An Optimal Feature Set With LBP for Leaf Image Classification

Plants are the integral part of human life. The preparation of Ayurvedic medicine from medicinal plants began thousands of years ago, and most of the people depend on Ayurveda for the recovery of diseases. The awareness of plants from manuscript, ancestors and Botanist helps to protect the plant which maintain the balanced ecosystem. Identifying plants through wet lab experiment is a time consuming task. The advancements in computational techniques have given considerable contributions in plant identification domain and it is gradually becoming an alternative method. The proposed method gives an optimal feature set which has evolved from LBP, GLCM, and HOG features extraction techniques. The combined feature vector is optimized with the help of Neighborhood Component Analysis (NCA). The feature selection and dimensionality reduction techniques have helped to enhance the classification performance and computational efficiency. Three plant datasets - Flavia, D-Leaf and Swedish Leaves - are used in the experiments for evaluating the proposal. Experimental result shows that proposed method has an average classification accuracy of 97.63% in 291.24 seconds computational time.