Concealment of persons using face recognition in circumstantial video image

Abstract

A circumstantial video image should convey sufficient information about the situation, while protecting the privacy of the specific person's in the scene. This paper describes a concealment system which automatically identifies a person by using face recognition, tracks him or her, and displays his or her image in a modified form such as in silhouette with or without name, or displays only a name instead of the original video image. A method for comparing situational information obtained from the modified or raw circumstantial video image with one required in several applications is also described. A subjective evaluation was carried out in order to ascertain how people evaluate modified video images from the observer or subject viewpoint. As a result of an experiment in which the several applications were considered, the silhouette with name list type of concealment seems to be the most appropriate observer-subject compromise.