Recently in Korea, CCTV is commonly seen on crowded streets. As of 2020, the number of CCTV installed by administrative agencies alone amounted to 1,336,653, which is increasing every year compared to the previous year [1]. Although the installation of CCTV is contributing to the operation of a safe city, the controversy over invasion of people's privacy is growing. In particular, since a human face can be used as personal identification information by itself, de-identification processing is required to protect personal information. Since the traditional de-identification method damages image information, although privacy is improved, there is a problem of low utilization.

In this paper, we propose a face image de-identification technique to solve this problem by using the adversarial attack method[2]. In the proposed method, Although human cannot identify the specific person of the original from the changed image but the face recognition model can increase the utilization of the image by allowing the person to identify the specific person of the original from the changed image.