Hand Vein Recognition Based on Feature Coding

A novel feature coding scheme based on back propagation neural network (BP) is proposed in this paper for accurate hand dorsal vein recognition. Feature vector is converted to a binary sequence, which can improve the performance of classification. Partition local binary pattern (PLBP) is extracted as the input of BP encoder and orthogonal Gold code is selected as the output code for BP encoder. Thanks to the orthogonal characteristic, Gold code can decrease relevance between different classes while enhancing the relevance within the same classes. Besides single-encoder by BP, the error correcting coding (ECC) is adopted in the combination-encoder to reduce the rate of error codes. Correlation classifier is taken as the final classifier. Experimental results show that feature coding strategy by BP combination-encoder achieves a high recognition rate of 97.60%.