基于可撤销人脸的模糊保险箱算法研究与实现

Research and Implementation of Fuzzy Vault Based on Cancelable Face

针对传统人脸特征识别系统安全性问题,提出了一种可撤销人脸的模糊保险箱算法。解决了由于模糊保险箱方案中存在真实细节点信息,导致人脸特征不安全的问题。在进行特征模板加密过程前,先用正交随机矩阵加密,加密后的模板具有可撤销性,不存储真实细节点信息,增强了系统的安全性。并在解密过程中的解码阶段,用Berlekamp–Welch解码算法代替CRC解码,该算法在解密阶段仅需根据解码集重构一次多项式,提高了算法的效率。改进后的模糊保险箱算法在生物特征模板受到攻击或者泄露时,能够随时删除并重新产生新的正交随机矩阵,并生成新的生物特征模板,具有可撤销性和安全性。该方案基本流程分为四部分:2DGabor-PCA特征提取、特征可撤销变换、模板加密、模板解密。在ORL人脸库进行测试,最佳识别准确率达到96%,经过对比实验验证了该方法的有效性,能够满足生物模板保护方案所需的不可逆性和可撤销性。

Due to the security problem of traditional face recognition system, we propose a fuzzy vault algorithm for cancelable face, which solves the problem that the face features are not safe because of the existence of real minutiae information in the fuzzy vault scheme. Before the process of feature template encryption, orthogonal random matrix is used to encrypt. The encrypted template is revocable and does not store the real node information, which enhances the security of the system. In the decoding stage of the decryption, Berlekamp-Welch decoding algorithm is used to replace CRC decoding. In the decryption, the algorithm only needs to reconstruct the polynomial once according to the decoding set, which improves the efficiency of the algorithm. When the biometric template is attacked or leaked, the improved fuzzy safe algorithm can delete and regenerate a new orthogonal random matrix at any time, and generate a new biometric template, which is revocable and secure. The basic process of the scheme is divided into four parts: 2 DGabor-PCA feature extraction, feature revocable transformation, template encryption and template decryption. In ORL face database, the best recognition accuracy is 96%. The experimental results show that the proposed method is effective and can meet the irreversibility and revocability of the biological template protection scheme.