Research On The Key Technologies Of Palm Vein Recognition

In today's information society,the demand for identity verification is becoming increasingly urgent.Traditional user ID,IC card,digital certificate and other authentication technologies have many defects and deficiencies in practical applications,especially the problems of inconvenient use,easy loss and forgery,which cannot meet the need of the current information society.The identification of unique biometric features carried by the human body,such as fingerprints,human faces,iris and DNA,has become a hot topic in domestic and foreign research.Palm vein recognition is a kind of biometric technology in the human body.It uses infrared radiation to acquire palm vein texture and uses special algorithms to extract features and use this feature as user identification technology.Compared to the widely used fingerprint identification,face recognition and other technologies,palm vein identification technology has an irreplaceable technical advantage.Because vein texture is in the skin tissue,texture is complex and the naked eye is not visible,there is no texture residue after touching objects,so it has strong concealment.At the same time,from the characteristics of the vein,we can see that only when human hands are alive,can the texture features be acquired by infrared acquisition device.And the vein texture created by drawing and 3D printing cannot be recognized by equipment,so it has the characteristics of natural living detection.In all kinds of human body biometrics,the palm vein recognition technology has a good application prospect by considering the factors of technology stability,user convenience,user acceptance and equipment cost.However,researchers at home and abroad to study the palm vein recognition is not comprehensive,the algorithm has a certain application prospect,but mainly is the identification and treatment of vein texture itself,few multi-modal Biometric Fusion,correlation analysis,sparse encoding and adaptive filtering technology and the palm vein recognition technology combined.This dissertation focuses on the key issues of palm vein identification technology in the research on image acquisition and enhancement,adaptive local texture extraction,multi modal biometric information fusion,correlation analysis and sparse encoding and other aspects of the optimization algorithm and strategy.The main work and innovation of this dissertation can be summarized as follows:(1)Study the ROI and preprocessing methods suitable for the feature extraction of palmar veins.One of the prerequisites for the effective work of the palm vein feature recognition system is the accurate ROI clipping and image preprocessing of the original image data.The selection of ROI location and size is of great importance for subsequent feature extraction and comparison steps.It has a profound impact on the recognition rate of the system.This dissertation proposes an algorithm that maximizes the ROI clipping area.With this method and the subsequent CLAHE gray level processing algorithm,we can get the best palm vein original image.(2)Study the adaptive local texture extraction method based on Gabor filter.Gabor features can be used to express image texture information and is widely used in the field of human biometric information recognition.In palm vein feature recognition technology,only the central frequency,main direction and standard variance parameters of Gabor filter match with local texture,it can achieve the best effect.For this reason,this dissertation proposes an algorithm of adaptively selecting local Gabor parameters according to the texture characteristics of different regions of the image,so that the vein feature can be extracted optimally.(3)Study the multimodal biometric information fusion based on correlation analysis.Assuming that the same user keeps sampling information in multi-modal biometric information base,the ideal case is that the sample should have strong correlation with the user and no correlation with other users' samples.Therefore,a mathematical method can be used to map the intra class distribution matrix and the inter class distribution matrix,so that the intra class correlation is the largest and the inter class correlation is minimal.In this dissertation,a multiset generalized canonical discriminant projection(MGCDP)method is proposed.This method first distinguishes single mode sample classification,mapping the intra class distribution matrix after reducing dimension,making the single mode intra class covariance matrix becomes diagonal matrix,and achieves the purpose that each sample is related to its class only.Then,in order to ensure that only the same users in the multimodal have correlation,the mapping method is used to transform the multi-modal feature matrix of users,so that the covariance matrix in the multimodal inter class is also a diagonal matrix.On this basis,serial multi generalized correlation discriminant project strategy(S-MGCDP)and parallel multi generalized correlation discriminant project strategy(P-MGCDP)are proposed,which achieve more than two kinds of biometric modal fusions.(4)The study of palm vein features representation based on directional multi-scale sparse coding method.In order to further improve the recognition accuracy and operational efficiency of the recognition system,this dissertation proposes a directional multi-scale sparse coding method,which processes the global and local range feature separately.The hamming distance is a measure of the distance between classes.The three main steps in this method are as follows.First is the direction discrimination step.In this step,the ROI region is preprocessed and the independent sub region is divided,and then the global and local direction information are obtained by Gaussian-Radon transformation.Then the Gabor filter and the sparse coding step.In this step,the multi-scale directional filtering operation of each sub region is performed,and the output results are coded sparsely.Finally,in the step of contour coding,the feature matrix is obtained by converting the compound feature of the specific direction to the binary feature vector.(5)Study the palm vein recognition method of local Gabor histogram fusion.In order to make full use of the information of the phase and amplitude of Gabor filter,and reduce the sensitivity of different images in the same vein registered users to brightness and slight displacement,this dissertation proposes the Fusion of Local Gabor Histogram(FLGH)method.In this method,the real and imaginary part of Gabor filter is extracted and fused by Local Gabor Principal Differences Pattern(LGPDP)and Local Gabor XOR Pattern(LGXP).The Fisher linear discriminant algorithm based on spatial domain partitioning is used to reduce the feature dimension of the palm vein and reduce the number of samples required for training.(6)Study the local texture of the palm vein feature.Palm vein can be regarded as a feature pattern with varying texture and structure.Therefore,local texture can be used to extract these features.The MB-LDP method proposed in this dissertation can get multi-scale texture in the variable center and neighborhood,extract the higher-order features with local derivative method,and encode the local spatial relations.Combining ROI sub regional division and the PCA dimension reduction technology,the multi-scale high order texture information is obtained under the reasonable feature dimension.The above results can be summarized as:for palm vein identification technology,by improving the key link of preprocessing,local feature extraction,information fusion and sparse encoding,so that the reliability of the recognition system is improved and provide theoretical support to reveal the key parameter and the recognition rate of system connection.