De-Identification of Facial Images Using Composites

May 10, 2011 3:20 - 5:00pm

Mark Edward Engelstad DDS, MD, MHI Oral and Maxillofacial Surgery, University of Minnesota
Genevieve Melton, MD, Surgery, University of Minnesota

Facial surgeons rely on facial photography to illustrate findings and educate other surgeons. Unlike other parts of the body, photographs of the face are readily identifiable, yet surgeons have a need to accurately present facial features for educational purposes. Traditional methods of de-identification may not fully protect identity or may decrease educational image quality. We have developed a novel method of using facial composites to de-identify full-face images while leaving educationally relevant components of the facial appearance unaltered. Methods: 20 composite facial images were developed. 10 of these contained components of faces familiar to the student volunteers. Volunteers viewed the faces twice—first unprimed, then primed to the presence of composites. Volunteers then rated faces de-identified by different methods.

Results: When unaware that they were viewing composite images, none (0/120) of the composite images containing familiar test faces were either identified or rated as familiar. When primed to the presence of the familiar faces, the identification rate improved significantly (74/120 (62%), p<0.001). Results were similar no matter which portion of the familiar face (upper (66%), mid (54%), lower (66%)) was present. Throughout the study, volunteers rated all composites as realistic patient images and as more effective than traditional methods for de-identification, like black bars over the eyes.