Bimodal Biometrics Recognition Method Fused With Finger Features

With the development of informatization and networking,biometric technology has been widely applied due to its convenience and reliability.Since different biometric technologies have their own advantages and disadvantages,how to carry out multi-modal biometric research and improve the reliability and security of biometrics has become the focus of research.This paper focuses on the fingerprint and finger vein dual-mode biometric technology,and the specific work is as follows:1.In view of the problem that low-quality images have greater influence on the biometric system,image preprocessing has been carried out.(1)A method for image quality assessment is proposed.Firstly,a reasonable image quality evaluation index is designed.Secondly,the improved k-means is used to divide cluster images into two categories: good images and bad images.Thirdly,the good image and bad image are trained by metricNet model,and the quality classification of all images is completed on this basis.Finally,the bad image is deleted under the premise of increasing the proportion of good image.(2)Histogram balance enhancement is carried out on the screened fingerprint image,and Clahe enhancement is carried out on the screened finger vein image,so as to provide a good input image for dual-mode biometrics.Experiments show that image quality classification has a high accuracy,which proves the effectiveness of this method.2.In view of the low robustness of traditional biometrics,the Vit-b/16 network is determined to be used for fingerprint and finger vein biometrics.(1)Select the classic ResNet34,Goog LeNet and Vit-b/16 networks to conduct biometric comparison experiments,and make a systematic analysis of the influence of Vit-b/16 structure and parameters on the experimental results.(2)Improve the Vit-b/16 by blocks,and design a parallel Vit blocks,changing the two connected blocks from serial to parallel.Experiments show that the performance of the improved Vit-b/16 network is stable and the recognition effect is better.3.Aiming at the low security and limited reliability of single-mode biometrics,a dualmode biometrics system based on Vit-b/16 was designed.(1)Based on the SE module,the feature fusion method of Vit-b/16 adaptive weight is designed,and a dual-mode biometric identification system is built,which improves the role of useful features in biometric identification.(2)Using a supervised contrastive loss function to train a multimodal network to solve the recognition problem of intra-class diversity and interclass similarity scenes,and improve the performance of dual-modal biometric recognition systems.Experiments show that the dual-modal biometric system can effectively improve the accuracy,security and reliability of biometrics,and has better robustness and generalization ability.