人脸识别技术在门禁、视频监控等公共安全领域中的应用日益广泛，人脸特征数据的安全性和隐私性问题成为备受关注的焦点。提出一种基于SIFT的人脸特征安全保护新算法，首次将随机投影应用到对人脸特征数据的保护中。该算法首先利用SIFT特征对旋转、尺度缩放、光照变化等具有较好稳定性的特点，提取有较好鲁棒性的人脸数据；然后根据用户密钥对SIFT特征进行不可逆变换，生成具有可重建性的人脸特征模板数据，从而实现对人脸特征数据的保护。实验表明，该算法在CMU、AR和Feret人脸数据库中均取得较高的识别率，不仅对人脸特征数据具有保护作用，并且对姿势、遮挡和表情的变化具有较高的鲁棒性。

With the growing use of face recognition in security and video control domain, there are growing concern about security and privacy of biometrics data. This paper proposes a security algorithm about face feature, which bases on the SIFT feature and random projection. Because the SIFT features are invariant to image rotation, scale and change in illumination，feature extraction is first performed on face images by SIFT algorithm. The SIFT features are transformed using invertible transformation which is generated by user specific key. All the above successive procedures produce the cancelable and non-invertible template feature data, which can achieve the protection of face data. In extensive experiments with publicly available face datasets CMU, AR and Feret, higher recognition accuracy is reached, which demonstrates that the proposed approach is not only able to protect the face data, but also robust to various complex conditions, such as changes in the pose, occlusions and expression, etc.