Unimodal and multimodal biometric verification using cancelable iris and fingerprint templates

Abstract

Biometric-based recognition systems is gaining the popularity day by day and have overcome passive issues of traditional human authentication systems. However, security theft and privacy invasion are two passive issues that still persist in the effective deployment of biometric-based authentication systems. Compromise of biometric data can potentially lead to serious security violation as the user’s biometric trait cannot be changed. In order to prevent the invasion of biometric templates, it is desired to morph the original biometric template through non-invertible or irreversible transformation function. This transformed template is referred to as cancelable template and can be replaced or reissued in case of compromise. The problem still persists if a protected multi-biometric template gets compromised. Objective of this thesis is to address the mentioned concerns associated with template protection and investigate the template protection schemes for unimodal and multimodal biometric traits with large scale biometric data so that the matching can be accomplished in transformed domain without compromising the verification performance. In this dissertation, we consider two biometric traits (iris and fingerprint). We propose an efficient template protection scheme for both biometric modalities. Next, we utilize both protected templates for cancelable multibiometric verification system. Our iris template protection scheme uses IrisCodes derived form 1-D log Gabor filter with rotation-invariant mechanism. Next, consistent bit vector is derived by aligning IrisCodes of different samples of the same subject. The consistent-bit vector is divided into equal sized words to form a decimal vector. Then, a Look-up table mapping based transformation has been applied for cancelable iris template generation. In cancelable fingerprint template protection, we evaluate ridge features with reference to ridge co-ordinate system. The computed features are invariant to translation, scale, and rotation. These features are uniquely encoded using Cantor pairing function. Then, we apply random projection for cancelable fingerprint template generation. Our protected multimodal verification scheme utilizes scores evaluated from the protected modalities. Corresponding to each modality, we integrate the scores from different matchers/classifiers based on the novel mean-closure weighting (MCW) mechanism. The fused scores obtained from different matchers for each trait are then combined using rectangular area weighting (RAW) mechanism. This two-level score fusion method is incorporated for cancelable multi-biometric verification.Finally, we proceed in the direction of hybrid fusion integrating both score and decision level mechanism to overcome the limitations related to unibiometric, multibiometric, and existing protected multibiometric systems. In order to perform hybrid fusion, we utilize our previous works to derive cancelable template for iris and fingerprint. Next, we apply the novel MCW to combine scores obtained from individual matchers. Then, Dempster- Shafer (DS) theory of evidence is employed to combine the decisions provided by individual matchers. The major contributions of the thesis are cancelable template generation for iris and fingerprint biometric modalities. The proposed template protection schemes preserve the desired criteria of irreversibility, revocability, and diversity of the cancelable transformation. Also, experiments performed on different databases confirm the potential robustness of the proposed transformation. Further, the proposed multimodal cancelable multimodal biometric verification techniques are able to attain performance improvement and provides adequate security to protect original biometric data. Thus, the proposed cancelable multibiometric verification proves to be effective for secure and accurate authentication.