Font Size: a A A

Research On Key Issues Of Multi-Modal Biometric Verification Based On Finger

Posted on:2015-08-01Degree:DoctorType:Dissertation
Country:ChinaCandidate:J L PengFull Text:PDF
GTID:1228330422992447Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
With the rapid development of information and network technology, the demands of personal biometric verification are increasing in the current social life. However, the unimodal biometric technologies result in some deficiencies, such as anti-spoofing, dis-crimination etc. On the other hand, multi-modal biometric technologies are the fusion of multiple biometric sources, which can offer higher security, better discrimination etc. Multi-modal biometrics has become the main trend in the field of biometric research. From the security point of view, this dissertation proposes the finger multi-modal biomet-ric verification scheme, which combines finger vein, fingerprint, finger knuckle print and finger shape features to meet the actual demand of acquisition and further improves the verification performance. In addition, the finger multi-modal template protection scheme is proposed to secure finger multi-modal feature templates. Therefore, the research on the key issues of finger multi-modal biometric verification has both academic value and practical significance.From the analysis of finger multi-modal biometric verification problems, this dis-sertation pays attention to finger multi-modal biometric verification under the following major topics:Firstly, for the low quality images related to finger vein verification, the preprocessing and image quality evaluation approaches of finger vein images are pro-posed. Secondly, to full use of the anti-spoofing and discrimination of finger vein traits, the feature extraction approach on finger vein images are proposed. Thirdly, the finger multi-modal biometric verification at feature level fusion is proposed to combine finger vein, fingerprint, finger knuckle print and finger shape features. The proposed feature level fusion approach can show the advantage of finger multi-modal biometric verifica-tion effectively. Finally, the corresponding finger multi-modal template protection scheme is proposed to secure the fused feature templates in finger multi-modal biometric verifi-cation. The relationship between security and verification performance of finger multi-modal template protection scheme is also analyzed in this dissertation.The main contributions of this dissertation are described as follow:(1) The preprocessing and quality evaluation approaches of finger vein images are proposed. For finger imperfect placement and uneven illumination in finger vein images, Region of Interest (ROI) location based on the physiological structure of human finger is proposed. The gradient, image contrast, and information entropy of finger vein image ROIs are extracted as the quality evaluation scores. The score fusion approach based on Triangular norm is proposed to evaluate ROI quality. The proposed evaluation method is more accurate than other existing methods in terms of the quality of finger vein images, and reduces the influence of low quality finger vein images on verification performance.(2) The finger vein feature extraction method based on Gabor wavelet and Local Binary Pattern (LBP) is proposed. This method utilizes LBP operator to describe the multi-scale and multi-orientation Gabor Magnitude Pattern of finger vein image ROI as GLBP feature. During the matching procedure of GLBP features, the Block-based Lin-ear Discriminant Analysis (BLDA) method is proposed to improve the discrimination of GLBP features. The proposed finger vein feature extraction approach has better verifica-tion performance compared to other existing feature extraction approaches.(3) Linear Discriminant Multi-set Canonical Correlation Analysis (LDMCCA) is proposed to fuse finger vein, fingerprint, finger knuckle print and finger shape features. The proposed LDMCCA approach solves the problem of fusing more than two modal bio-metric features in the existing feature level fusion approaches. It can not only preserve the intrinsic discrimination between within-class features and between-class features, but also reduce the dimension of fused features by the canonical correlation relationship among multi-modal features and improve the discrimination of fused features. The proposed finger multi-modal verification approach based on LDMCCA feature fusion has made significant performance improvements over finger unimodal verification approaches and other existing finger multi-modal verification approaches at feature level or score level fusion.(4) The analysis of security and verification performance on finger multi-modal tem-plate protection is proposed. According to the characteristics of proposed multi-modal feature level fusion approach, Fuzzy Commitment Scheme (FCS) is utilized to secure the finger multi-modal templates. From the security analysis of FCS, the objective function is also proposed to select the error correcting capability of Error Correction Code (EC-C). Although there is a contradictory relationship between verification performance and template security, it can be made the tradeoff between them by choosing the rational cor-recting capacity of ECC. The proposed finger multi-modal template protection approach outperforms finger unimodal and score level fusion counterparts in terms of verification performance and template security.
Keywords/Search Tags:Multi-modal biometrics, Finger, Feature fusion, Template protection, Secu-rity analysis
PDF Full Text Request
Related items