An approach to the de-identification of faces in different poses

Publicly captured surveillance videos and images serve as a rich source of biometric identifiers. Of these, the face is the one most frequently used for the identification of people. In order to protect a person's identity, whenever it is not absolutely required, the face should be de-identified. One of the problems that any naturalness-preserving face de-identification method should address is the variation in the face pose. In this paper we build several active appearance face models on the database of face images and then use them to detect face pose as well as a gallery of faces that is then used for the de-identification. The results can be seen as a first step to the face de-identification in video.