Dorsal Hand Vein Recognition Algorithm Based on Improved Nonnegative Matrix Factorization

To enhance feature effectiveness during hand dorsal vein recognition, we propose a new recognition algorithm based on improved nonnegative matrix factorization (NMF). Firstly, after dividing vein image into blocks, we use the mean and the average gradient amplitude of sub-image as image original features. Secondly, we apply NMF in the feature matrix which is formed by combining the original feature vectors of all training samples, where the coefficient vectors are imposed by sparse and discriminant constraints, and the improved NMF model can be acquired. Thirdly, we use a projected gradient method to solve the NMF model, and new feature basis and feature vectors are obtained. Finally, new feature vectors are classified by K-nearest neighbour (KNN), and the vein object is identified successfully. Experiment results show that the proposed algorithm has high correct recognition rate and good real-time performance.