FIS: Facial Information Segmentation for Video Redaction

摘要

The camera is a ubiquitous and inseparable part of our daily life. However, the camera also puts people's privacy at risk because people are more likely to be unexpectedly captured due to the increased amount of cameras. Video redaction systems have been proposed to selectively redact facial information in a video. However, they are either not secure enough to protect all facial information or too secure to redact image areas not revealing facial information. To solve this problem, this paper proposes the Facial Information Segmentation algorithm (FIS) that combines the Harris Corner, the color information and an off-the-shelf face detection algorithm to identify pixels revealing facial information. We evaluate this method by comparing it with the human trace tracking (HTT) and an off-the-shelf face detection algorithm (FD) proposed in earlier works. The result demonstrates that FD is unsuitable for video redaction. Further, compared with HTT, FIS achieves higher background preservation with negligible loss of video privacy in most cases.