Font Size: a A A

Study Of Several Key Problems In Biometrics

Posted on:2019-02-10Degree:DoctorType:Dissertation
Country:ChinaCandidate:X Z LiFull Text:PDF
GTID:1368330545453580Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Biometrics is a novel identity authentication technique by using biometric of human.Due to its security and convenience,biometrics is applied widely in the area of finance,security and so on.In recent years,fingerprint,face,gait and finger vein are widely used as biometrics.The biometrics system mainly includes preprocessing,feature extraction,matching and multimodal fusion.In this thesis,we focus on four problems:(1)the features used in existing finger vein recognition ignore the difference of users and discriminative information;(2)the features used in existing finger vein retrieval methods are typically single features and the used information is not comprehensive;(3)it's difficult for existing fingerprint segmentation methods to deal with intensity inhomogeneity problem very well;(4)how to fuse multimodal information effectively in high security application.The main contributions can be summarized as follows:1.In order to improve the user difference and discrimination of features in finger vein recognition,finger vein recognition based on Personalized Discriminative Bit Map(PDBM)is proposed.Local Binary Code(LBP)is firstly used to extract the binary code,and then the personalized consistence rule is developed to learn the personalized stable bits from LBP code.The obtained personalized stable bits can reflect the main vein characteristics of each user,which can capture the user difference.In order to capture more discriminative information of features,discriminative rule is developed to select the discriminative bits from the personalized stable bits,generating PDBM.Experimental results show that PDBM not only obtains higher accuracy,but also improves matching speed.2.In order to solve the features used in existing finger vein retrieval methods are typically single features and the used information is not comprehensive,finger vein retrieval based on multi-scale fusion quantization encoding is proposed.Firstly,Multi-scale Texture Features,Multi-scale Intensity Features and Multi-scale Shape Features are fused to capture global and local intensity,texture and edge characteristics of finger vein.Then,Principal Component Analysis(PCA)is employed to compress the features.After that,Iterative Quantization method encodes the compressed features into similarity-preserving binary codes.Finally,we obtain the candidate images through hamming distance matching methods.Experimental results demonstrate that the proposed method can improve the retrieval accuracy.3.To solve the intensity inhomogeneity problem,fingerprint segmentation method based on vision perception is proposed.The local features which contain local intensity contrast,gradient and texture are extracted firstly for each pixel.And then,improved Fuzzy-ART algorithm is proposed for pixels classification.For the proposed method,the local features can capture the important local information which can address the problem of intensity inhomogeneity.Experimental results demonstrate the effectiveness of proposed method.4.In order to make the biometric recognition systems suitable for high security application,a systematic analysis of the special performance requirements of high security applications and define double low problem were proposed.The proposed hybrid ensemble framework is a general solution that can be applied to the ensemble of various biometric verification algorithms of a biometric system.By setting a sufficiently high threshold to guarantee zero or close to zero FAR and using our framework to reduce the FRR as much as possible,we can see strong results on three representative tasks exhibiting the hybrid ensemble frameworks strength.The proposed hybrid approach significantly outperforms traditional verification methods.This thesis analyzes four problems:1)lack of discrimination of features;2)single features information is not comprehensive;3)intensity inhomogeneity problem of segmentation;4)effective multimodal fusion in high security application.The solutions to these problems are proposed,i.e,Finger Vein Recognition based on Personalized Discriminative Bit Map,Finger Vein Retrieval based on Multi-scale Fusion Quantization Encoding Features,An Improved Fingerprint Image Segmentation Algorithm based on Visual Perception Model and A Hybrid Biometric Identification Framework for High Security Applications.The development of this study enriches the techniques of preprocess,feature extraction and multimodal fusion in biometrics,and further improve the performance of the biometrics systems.
Keywords/Search Tags:Biometrics, Personalized Discriminative Bit Map, Finger Vein Retrieval, Fingerprint Segmentation, Hybrid Fusion of Multimodal Biometrics
PDF Full Text Request
Related items