Research On Personal Identity Recognition Method Based On Multi-Biometric

Nowadays, with the rapid development of scientific and technological information, the social security and information security become more important than ever. As the indispensable premise for security, personal identity recognition is a crucial and social problem and hot issue in the academic research. With the advantage of reliablility, stability and validity, biometric recognition has been developing rapidly. However, it has been found that the single modal biometric shows some inherent drawbacks in accuracy, universality and anti-counterfeit. This problem can be addressed by installing multiple sensors that capture different biometric traits, known as multimodal biometric. At present, the multimodal biometric techenology is still on the exploratory state and need future research.A multimodal biometric scheme based on hand vein, iris and fingerprint was proposed in this dissertation, which could solve the critical technical issues in personal identity authentication. The alogorithm of feature extraction and feature matching for single biometric was proposed for improving the accuracy and real-time. Matching scores normalization methods had been researched. Fusion recognition algorithm for multimodal biometric at matching score level based on classic fusion theory, Bayesian decision theory and kNN-SVM theory were developed separately. The experimental results show that the proposed fusion recognition algorithm has high accuracy. Finally, we designed an experimental multimodal biometric recognition system, and preliminarily studied the recognition realization in DSP.The major innovations of the dissertation are summarized as follows:1. A fast recognition algorithm of hand vein using SURF descriptors is proposed. The hand vein image scale-space is constructed throuth three octaves expansion, and then the local SURF features are extracted and matched based on the Euclid distance. This algorithm is not sensitive to hand shift and rotation, which improves the accuracy and reduces the computing time greatly.2. The matching score normalization model is established and classic fusion methods is employed for multimodal biometric authentication based on hand vein, iris and fingerprint. Matching scores of hand vein, iris and fingerprint are normalized, and effect of normalization methods on fusion recognition performance are analyzed by using of discrimination parameter. The matching scores data fusion is realized by using classic fusion method at matching score level, and personal identity authentication result is finally achieved.3. A fusion authentication algorithm based on Bayesian decision theory for multimodal biometric traits of hand vein, iris and fingerprint is proposed. Take the matching scores of three biometric traits as input of Bayesian classifier, and personal identity authentication result is achieved by Bayesian decision function.4. A fusion recognition algorithm based on kNN-SVM theory for multimodal biometric traits of hand vein, iris and fingerprint is proposed. Firstly, hand vein feature matching is processed based on k-NN method to achieve identity preliminary recognition. Then the identity range is limited to k, and final identity result is decided based on SVM method using iris and fingerprint traits.5. Multimodal biometric authentication implementation in embedded system is achieved. The experimental system is designed based on DSP and FPGA for multimodal biometric authentication using hand vein, iris and fingerprint. The proposed algorithm transplantation from PC to the DSP-based system is preliminarily studied. A simple recognition experiment in DSP-based system is achieved finally.