Securing fingerprint images using data hiding techniques

Abstract

Rapid growth in technology and digitisation has enabled person authenti cation from a remote location. Compared to password based authentica tion, biometrics system gained popularity due to security and uniqueness. However, they su er from false matching and false non-matching problem which attacker can exploit e.g., attacking database to gain illegal access. Securing database provides immunity against unauthorise data injection, tampering and it also gives privacy protection. Watermarking provides an e ective solution against manipulations on the database. The biomet ric templates or images are watermarked and authenticated before their delivery to underlying biometric system. The conjunction of watermarking and cryptography enhances security with privacy protection. The proposed research work is focused on a scheme which provides privacy protection and immunity against unau thorise data injection and tampering. The research work use ngerprint images for development and testing of algorithms. The initial work develops semi fragile watermarking method to authenti cate multiple vulnerable points of the ngerprint system. Though it pro tects multiple points but irreversibility of watermark a ects underlying biometric system accuracy. This motivates development of authentica tion mechanism based on reversible watermarking. The improved method secures multiple vulnerable points without changing receiver operating characteristics (ROC) of the ngerprint system. However, this does not solve privacy protection problem as watermark is embedded in the plain ngerprint image. This led to development of the scheme which is commutative between re versible watermarking and key based random permutation. It has equiv alency in following operations. E(W(IKW)KE) = W(E(IKE)KW) where, W, E are watermarking and encryption functions while KW, KE are watermarking and encryption key respectively and I is the ngerprint image. The commutative scheme secures multiple vulnerable points with privacy protection. However on detailed security analysis, the method is found susceptible to unauthorise data injection and Denial of Service (DoS) attacks through copy attack and Most Signi cant Bit (MSB) tampering respectively. This necessi tates use of the image dependent watermark along with commutativity to improve the method. Results of the improved commutative method showcase that it is resistive to copy attack and MSB tampering. Finally, the thesis proposes novel idea of compression friendly privacy protection and authentication for ngerprint image. The method protects privacy sensitive region (PSR) of a ngerprint by lling it with uniform valued pixels. The PSR is losslessly compressed and reversibly embedded into privacy protected image. Uniform value PSR reduces entropy of the image making it compression friendly and reversible watermarking provides authentication.