A Face Feature Encryption Algorithm Based On LNG-score And Fuzzy Commitment

With the rapid development of information technology, information security has become a major challenge of the global information. Cryptography is the basis of information security, and the security of the key is the dependency of cryptography system. The keys are linked to the user identity by means of passwords or tokens. However, the passwords or tokens are assigned secrets and are not strongly associated with the person’s identity. Therefore, these secrets can be stolen or lost and cryptographic techniques cannot detect the theft. Fortunately, biometrics and cryptography can be combined to design better and more secure systems. Such systems are denoted as biometric cryptosystems. The strong association of biometric characteristics with the user’s identity can be utilized to provide the trust required in cryptography. Moreover, the cryptographic techniques can be employed to provide protection to the biometric data without compromising privacy. The biggest difficulty in biometric cryptosystems is that cryptography is precise whereas biometric data is fuzzy. Fuzzy commitment is the classic one of the options to resolve that contradiction.Firstly, theselection algorithm of face feature is initially researched and then optimized in this thesis. LNG-score algorithm is the combination of normalized NG-score algorithm and sensitivity analysis algorithm. Unrelated or high-sensitivity features can be effectively eliminated by this algorithm, thus the anti-interference performance is improved efficiently, and a solid foundation for the realization of fuzzy commitment algorithm is laid.Secondly, encryption algorithm of face feature which is based on LNG-score and fuzzy commitment is researched. This algorithm includes two processes: encryption process and decryption process. As to encryption process, XOR algorithm is conducted on random key. Its codes have been corrected and face feature has been through LNG-score algorithm. Then its value is stored in smart card. Meanwhile, hash value of random key is also stored and then random key would be destroyed. During the process of decryption, similar face features should be provided so as to decode the key which is randomly generated during encryption process. There is a string of character string which has error-correcting capability stored in smart card. Therefore, face feature information and privacy of users can be effectively protected.Finally, an experiment based on ORL face feature library and YALE face feature library is conducted. Result shows that the algorithm in this thesis can effectively separate true users from fake users with a high accuracy. Moreover, it has low FRR, FAR and the whole performance of this system is improved.