Research On Multimodal Biometric Identification Based On Fingerprint And Finger Vein

With the continuous development of science and technology,information security has become more and more critical in the daily life of contemporary people.Biometric recognition technology is increasingly valued for its stability,reliability,and security.The use of single biometric identification is vulnerable to deception and forgery,so there are some limitations in practical applications.Multimodal biometric fusion recognition can make up for the shortcomings of single biometric recognition to a certain extent due to the fusion of different single-modal biometric features.With the rise of deep learning technology,many researchers have applied it to multimodal biometric fusion recognition and achieved certain results.In order to further improve the recognition accuracy,this thesis also studies multimodal biometric fusion recognition based on deep learning.The main work is as follows.(1)In the existing multimodal biometric fusion recognition research,the discriminative features of a single modality are not fully extracted,and the complementary information between different modalities is not fully and dynamically used for fusion recognition,resulting in insufficient final recognition accuracy.To solve this problem,this thesis proposes a multimodal fusion recognition algorithm based on the channel spatial attention mechanism.Firstly,a dual-channel parallel finger single-modal recognition network is designed to extract fingerprint and finger vein features.When the network extracts features,it is not affected by other features.Then,aiming at the problem of inconsistent feature sizes of fingerprint and finger vein extracted,adaptive pooling is used to unify the sizes of fingerprint and finger vein features.Finally,a channel-spatial attention multi-feature fusion module is designed to fuse fingerprint and finger vein features.The final recognition result is obtained by the Softmax function.The simulation results on three-finger multimodal datasets show that the proposed algorithm achieves the highest recognition performance compared with some existing algorithms.(2)In practical application scenarios,due to the limitation of computing resources and memory capacity of application devices,the existing multimodal biometric identification systems are difficult to be deployed on some embedded devices or mobile devices.To solve this problem,this thesis proposes a lightweight multimodal fusion recognition algorithm based on dynamic channel exchanging.Firstly,a lightweight finger multimodal recognition network is constructed by referring to the deep separable convolution and the existing lightweight network model.Then,shared convolution,pooling and activation function,private Batch Normalization(BN),and dynamic channel exchange are used to realize the communication and fusion between different modes of fingers.Finally,the L1 norm penalty is introduced to facilitate communication and fusion between different modes.The simulation results on three-finger multimodal datasets show that the proposed algorithm can effectively reduce the number of parameters,computation,and storage space required by the network model to ensure certain recognition performance.