Performance Evaluation Of Convolutional Neural Network In Palmprint And Palm Vein Recognition

In today's information and network society,the technology of biometrics has become one of the most effective solutions for personal authentication.Biometrics is a field of technology that uses automated methods for identifying or verifying a person based on a physiological or behavioral trait,which has shown significant advantages over traditional personal authentication mechanisms,such as keys,passwords,personal identification numbers,and smart cards.In recent years,two emerging biometrics technologies,palmprint recognition and palm vein recognition,have received much attention.It is wellknown that deep learning technology has been widely and successfully used in various fields.Particularly,deep learning has become the most important technology for pattern recognition.However,in the fields of palmprint recognition and palm vein recognition,deep learning technology has not been well investigated.In this thesis,we make a largescale performance evaluation on palmprint and palm vein recognition using the Convolutional Neural Network(CNN).The major works of this paper are as follows:(1)Introduce and analyze the classic models of CNN in details.First,the basic knowledge of neural network is introduced.Second,the structure of classic model and the relationship between them are elaborated.(2)Carry out large-scale performance evaluation of CNN on palmprint and palm vein recognition.On five palmprint databases and two palm vein databases,the recognition performances of nine network models are evaluated in the different conditions including different network structures,learning rates,numbers of layer and amounts of training data,etc.(3)Propose a CNN-based method for palmprint recognition and palm vein recognition.In this method,CLAHE is used to enhance image quality in database,and SPP-ResNet is designed in combination with the spatial pyramid pooling to solve the problem that CNN needs fixed size for input.For the existing Siamese Network palmprint verification method,modifications have been made in feature extraction and network training.The experimental results show that the large-scale evaluation experiment for palmprint and palm vein recognition can speed up and provide guidance for relevant research.SPPResNet based on CLAHE can receive input of various scales and improve the recognition rates.The modified Siamese Network verification has also achieved good results.