Research On Multi-modal Biometric Identification Method Based On Convolutional Neural Network

Biometric identification refers to the technology that uses the inherent physiological characteristics of human body to identify individuals through the close combination of computer,optical,acoustic,biosensor and biometric principles and other technological means.With the growing demand for information security,biometric recognition technology has been widely used in daily life,such as fingerprint and face recognition of smart phones,fingerprint recognition of access control system and palmprint and iris recognition of identity recognition.Due to the disadvantages of single-modal biometric recognition,such as poor security and low reliability,multi-modal biometric recognition has become the focus of the research due to its better robustness and generalization ability.In addition,in recent years,the emerging deep learning method has achieved good results in many recognition fields because it can carry out self-learning of features and realize high-level expression of features.Therefore,the research of multi-modal biometric recognition based on deep learning has important practical significance and application prospect.In view of the excellent performance of deep learning theory and method in various recognition tasks,this thesis proposed a multi-modal biometric recognition method based on three biological characteristics of face,iris and palmprint under the framework of deep learning.Firstly,in order to verify the effectiveness of the method of biometric recognition based on deep learning,we designed the different structure of Convolution Neural Network(CNN)model for single-modal biometric recognition.Then,in order to improve the performance of biometric recognition,combined with the results of single-modal recognition,we constructed a CNN model based on feature fusion for multi-modal biometric recognition.Finally,in order to explore the influence of feature fusion method and mechanism on recognition performance,we proposed a variety of feature fusion strategies,and an optimal network fusion structure through comparative experiments.In order to verify the feasibility and effectiveness of the proposed multi-modal biometric recognition method based on CNN,this thesis carried out a large number of experiments in three standard databases(CMU PIE,PolyU,CASIA V1.0),and thoroughly analyzed the influence of various CNN structures and parameters on the experimental results.The experimental results show that the multi-modal recognition method has higher recognition performance than the single-modal recognition method.In the multi-modal recognition method,the dual-feature fusion mechanism can further improve the multi-modal recognition performance and solve the disadvantages of high error rate and instability of single-modal,which has the good market prospect and application value.