参考来源：
To evaluate the protection provided by our constructions against inference attacks, we use the AT&T Faces data set [17]. It consists of a total of 400 tightly cropped face images, showing 10 different angles for each of 40 human subjects. Each image contains 92x112 grayscale pixels.
来源： Approximate Thumbnail Preserving Encryption
[17] Ferdinando S Samaria and Andy C Harter. 1994. Parameterisation of a stochastic model for human face identification. In Applications of Computer Vision, 1994., Proceedings of the Second IEEE Workshop on. IEEE, 138–142.
网址：
GTDLBench
数据库基本情况：
人脸数据：
