Research On Dual-Modal Biometric Identification Technology Based On Feature Layer Fusion

At present,single mode biometrics such as fingerprint identification,iris identification have many disadvantages,such as low accuracy,poor security,high limitations,and so on.Based on this,this paper designs and studies a kind of biometrics feature layer fusion recognition method,which has the advantages of high security,unduplicatable,in vivo detection,etc.The main contents of this subject are as follows:(1)In order to eliminate the interference of the background,noise and other useless information of the image in the open data set of the finger vein,the finger edge of the image is firstly detected and the false edges are removed;the image is rotated and rectified through the midline fitting,and the inner tangent lines of the upper and lower edges of the finger are found;and the location of the phalangeal joint is found by using the change of brightness in the horizontal direction to cut off the area of interest.CLAHE and 2D Gabor filter are combined to enhance the image of ROI and obtain clear finger vein information.(2)In view of the fact that the open digital vein data set is small in scale and the training of neural network is easy to cause the problem of over-fitting,a method combining data enhancement with the generation of confrontation network is proposed to generate digital vein data set.The VGG-19 binary classifier was used to select the complete venous image,and the image quality was evaluated by information entropy,energy gradient function and grey variance product function.The results show that the digital vein images generated by this method have the same distribution as the real ones,and the quality of the images is better than that of the real ones.(3)In order to make up for the defects of single biometric feature recognition,a method of dual-modal feature layer fusion recognition based on convolution neural network is proposed.Self-attention mechanism is adopted to update weights of finger vein features and face features;mixed features are formed through channel cascading;deeper features are extracted by convolution operation;residual structure is used to fuse them with finger vein features and face features to maximize effective feature information and avoid feature information loss in feature fusion module and form new features for classification and recognition.Experiments on public data sets and generated data sets show that the accuracy of dual-mode feature layer fusion is 98.84% at least and 99.98% at most.The research in this paper breaks the limitations of single biometrics,improves the accuracy and security of biometrics,provides a new scheme for the security of identity information,and has great significance for biometrics.