Attributes Preserving Face De-Identification

bin yan, mingtao pei, zhengang nie; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2019, pp. 0-0

In this paper, we propose a Face de-identification method to remove the identification information of a person while maintaining all the face attributes such as expression, age and gender. Motivated by the k-Same algorithm, our method consists of three steps: first, k face images are selected randomly. These k face images may contain same or different face attributes with the test face image. Secondly, ELEGANT model is employed to transfer attributes from the test face to the k selected faces. After attributes transferring, the k selected faces have the same attributes as the test face. Then we average the k selected faces as the de-identified image of the test face. Experimental results show that our method can de-identify a face image while preserving all of its attributes effectively.