Recognition Method Based On Finger Vein And Finger Knuckle Print

The rapid development of information industry technology put forward new requirements for identity authentication technology,which requires intelligence,convenience and security.Biometric recognition technology has become a hotspot research topic in the field of identity authentication.Single-mode biometric recognition technology still has limitations on application scenarios and security,in order to further improve its performance accuracy,multi-modal biometric fusion technology has received extensive attention.In this paper,finger vein and finger knuckle print as multi-modal fusion objects are researched.The main work is shown as follows:(1)Due to the complexity of traditional feature extraction methods related to finger vein and finger knuckle print and its high degree of human intervention,a feature extraction method based on convolution neural network is adopted.Secondly,to solve the problem that the sample size of finger vein and finger knuckle print is too small to support the convolutional neural network to start training with randomly initialized parameters.Therefore,the transfer learning method is adopted to obtain the network models.(2)This paper proposed a feature-level fusion method of finger vein and finger knuckle print.The network model optimized by single mode parameters is used as a feature extractor to extract feature vectors of finger vein and finger knuckle print.The two vectors are connected in series and classified by SVM classifier to realize feature level fusion of finger vein and finger knuckle print.The accuracy is improved by 0.34%~1.67% compared with the highest recognition rate of all kinds of single mode models.(3)This paper proposed a score-level fusion method of finger vein and finger knuckle print.Utilizing the network model with single modal parameters after optimized to classify finger veins and finger joints,the matching scores are obtained.After the optimal weighting ratio is decided with different weights factor,the final fractional fusion scheme is established.The accuracy is 0.33%~3.67% higher than that of the various single mode models.The paper has 40 figures,13 tables and 54 references.