Research On Palmprint Recognition Based On Deep Learning

Traditional personal identifications such as ID cards,passwords,etc.are easy to lose and forget,and the security is poor.On the other hand,the biometrics technology has high security,convenience and other advantages that attract wide attentions as an emerging biometric recognition technology,palmprint recognition has the advantages of high recognition accuracy,simple and convenient collection equipment,and good user acceptance.At present,palmprint recognition technology is still in the stage of theoretical exploration.Compared with other biological features such as fingerprints,faces,signatures,etc.,palmprint recognition technology has great development potential and application prospects.However,in the palmprint collection process,the palm posture and the degree of expansion and contraction may cause the palmprint image to have translation,rotation,scaling,distortion,etc.This phenomenon is called palmprint deformation.The existence of deformation will lead to a decrease in recognition accuracy,and the existing palmprint recognition algorithm cannot solve the above problems well.To tackle the above problems,this paper proposes to improve the recognition accuracy by using palmprint local texture information or local geometric features under deep learning architect.The main work of this paper is as follows:In this paper,the superiority of convolutional neural network in palmprint recognition is firstly proposed,and Gabor filter is introduced on the basis of convolutional neural network.During the training procedure of convolutional neural network,category labels are not needed,and the unsupervised training is achieved.We validated our algorithm on both the CASIA and IITD palmprint databases,and in most cases,our algorithm achieved higher recognition accuracy.What is more,in this paper,the four types of three-dimensional palmprint curvature feature extraction methods are described,e.g.,average curvature,Gaussian curvature,shape index and surface type.Then,the four different palmprint curvature features are combined with the current popular deep learning framework.Note that the three-dimensional palmprint recognition is mainly judged according to the palm-grain surface concave-convex structure,and thus is not easily affected by external factors.Therefore,in our proposed algorithm,the problem of palmprint deformation is overcome by extracting the three-dimensional palmprint curvature feature,and combined with the deep learning framework for recognition.We used the 3D palmprint dataset published by the Hong Kong Polytechnic University for evaluation.The effectiveness of our proposed algorithm is verified by four different convolutional neural networks,and it is far superior to the traditional palmprint recognition methods when considering running time.