Research On Related Techniques Of Palm Vein Recognition System

With the arrival of the information society,how to effectively protect our information and property security is gradually becoming a problem that everyone needs to solve.However,it's not easy for people to find an efficient way to protect their information security when there are frequent identity theft.The form of biometric identity authentication has gradually replaced traditional authentication methods and become people's first choice for safety.The palm vein as one of human body's biological characteristics,is unique,stable,hidden under the epidermis and has complex structure which is difficult to be copied.These features make the palm vein much safer than other biological characteristics.As a frontier subject in the field of biometric identification,the palm vein recognition technology has a wide range of application prospects due to its hig h security and other advantages,and has become a hot spot in recent years.The palm vein recognition includes three stages:preprocessing,feature extraction and classification recognition of palm images.In the stage of image preprocessing,the image in the palm image database acquired by the infrared device is processed in series,the excess palm portion is discarded and the vein pattern is extracted.So the complete palm vein pattern image is finally obtained;In the stage of extraction,block-based PCA,block-based 2DPCA and block-based KPCA algorithms were used to perform palm vein feature extraction studies respectively.Finally,experiments demonstrate that the block KPCA method has a higher recognition rate than linear block PCA and block 2DPCA.In the stage of classification and identification,the BP and RBF network were used to classify and identify the palm vein images respectively.Experiments show that the RBF neural network has faster convergence and shorter training time.The specific research content of this topic is as follows:(1)Research on the palm image preprocessing correlation technique.The palm image preprocessing includes image graying,binarization,morphological processing,extraction of palm contours,location and ROI extraction,ROI normalization,vein image denoising,vein image enhancement,vein pattern segmentation,and vein pattern refinement in total of 10 steps.This topic focuses on the segmentation algorithm of palm vein image on the basis of above preprocessing steps in traditional methods.For the small holes in the vein pattern which appear after the NiBlack processing and unnecessary non-vein parts in the image segmentation,the algorithm is improved.The combination of the NiBlack method and the region growing method has effectively improved the recognition rate.(2)Research on the feature extraction methods of palm vein recognition.The PCA and 2DPCA algorithm of linear subspace are combined with the idea of image segmentation.In the block PCA and block 2DPCA algorithms,each image is divided into several image sub-blocks of the same size and not overlapping with each other,which increased the number of samples.Considering that there are still a lot of non-linear information in the palm vein image,this topic applies the KPCA method to the research of the palm recognition,and adopts the block KPCA algorithm to extract features and improve the palm vein recognition rate.(3)Introduce BP neural network to classify and recognize palm images.The theory of BP network,algorithm and structure are studied.And the appropriate neural network structure and learning method is designed to train the network and achieve effective classification and identification.Finally,the applicability of the BP neural network in the palm vein recognition field is verified by experiments.(4)Introduce the RBF neural network to classify and recognize the palm images.The learning process of RBF neural network and the data processing methods of each layer are studied.Combined with the topic of the objectives,the appropriate learning method and activation function is selected.By constantly modifying the network to achieve convergence,the parameters are obtained.Finally,the RBF neural network is compared with the BP neural network.Experiments show that the RBF neural network has faster convergence,shorter training time,and can improve the recognition rate of the block 2DPCA algorithm.