Empathy through Aesthetics: Using AI Stylization for Visual Anonymization of Interview Videos

Faces are the primary source for identifying individuals, which need to be obstructed to achieve anonymization in images and videos. However, human faces are also one of the most important sources of emotional information, which can be lost during the standard anonymization techniques such as blacking out or blurring. The absence of the emotional signals from the face could have negative effects on the viewer in terms of the feeling of empathy towards the anonymized individual. In this paper, we present the use of AI-driven image stylization techniques as a solution for preserving the emotional facial cues in a video to evoke empathic behavior while ensuring anonymization. We show the effectiveness and shortcomings of our method by testing it against both standard anonymization and non-anonymized techniques. We conclude with next steps and future work to improve our technique.