Multi-modal Biometric Recognition Research Based On Fusion Algorithm And Deep Learning

With the rapid development of the Internet,threats to network security continue to emerge,people pay more attention to personal information security.As a common personal identity authentication method,biometric identification systems have been widely used in different fields:such as financial mobile payment,personal identity authentication,medical information management,and Internet of Things security.While receiving attention from the market,people have also raised higher requirements for their security,accuracy,and user acceptance of identification systems.Faced with these opportunities and challenges,recognition systems that only use a single feature have low security and are vulnerable to external attacks.In response to the shortcomings of single-modal biometric recognition proposed above,the multi-modal biometric recognition systems break these limitations by fusing information from multiple modes.This paper introduced the application status of multi-modal biometric recognition in the field of personal authentication in detail from financial and medical treatment.Based on the introduction of concepts such as iris normalization,vein enhancement,correlation analysis theory and convolutional neural network,a multi-modal biometric recognition system combining iris,palm vein and finger vein was innovatively proposed.By analyzing the application of different fusion levels in multi-modal recognition system,this paper proposed a new multi-modal biometric recognition algorithm based on multi-layer fusion strategy.The fusion strategy included an improved feature fusion algorithm and a novel weighted voting decision fusion algorithm.The former algorithm used the framework of multi-set parallel input to solve the limitation of input vectors in feature fusion.The latter algorithm guided the final decision through the distribution of feature matrix information to make up for the shortcomings of insufficient information in decision fusion.In this paper,the proposed recognition algorithm was tested simultaneously on three databases:CASIA,PolyU and SDU.Compared with other recognition algorithms,the proposed algorithm had higher recognition accuracy and stronger robustness.Experimental results showed that the average accuracy of this method was 99.33%,which was better than most other multi-modal biometric recognition algorithms.In view of the shortcoming of iris in multi-modal systems that needs to manually extract features,this paper proposed a multi-modal biometric recognition algorithm based on deep learning and score fusion.According to the model characteristics,the combination of iris recognition network and traditional vein recognition method was selected.Then the multi-modal recognition system was integrated with a score-level fusion algorithm.Though the research and analysis of deep learning,an improved convolutional neural network was used to automatically recognize iris images,which simplified the manual extraction of complex iris features.The convolutional neural network used the residual connection structure and the Inception module to improve its ability to express high-level features.It accelerated the network convergence by the ameliorated loss function,and improved the recognition performance of the network.Considering the strong directionality and regularity of vein images,traditional pattern recognition method was retained as preferred vein recognition method.Then score-level fusion strategy was used to obtain the final matching score.By combining the traditional pattern recognition methods and the emerging machine learning methods,multimodal biometric recognition systems took the best of both approaches.The systems mined deeper semantic information in images while ensured high recognition speed,and made the generalization ability of the multi-modal biometric recognition stronger.Experimental results showed that the multi-modal biometric recognition algorithm based on deep learning and score-level fusion proposed in this paper was faster than traditional recognition methods.