Face Recognition in Surveillance Video for Criminal Investigations: A Review

Abstract

Face recognition from surveillance video in a forensic scenario is a very challenging task. Much of the existing work focuses on face recognition in a video where the video frames are of high-resolution, containing faces in frontal pose and in optimal lighting conditions. However, new challenges are encountered as applications of face recognition advance from cooperative and constrained scenarios to uncooperative subjects in unconstrained scenarios such as video surveillance. These challenges are due to low image resolution, variant expressions, face orientations, partial occlusion, complex background and the differences in surrounding illumination. In criminal investigations, the aim is to identify a culprit by collecting the information from the various face input media that includes video tracks, still face images, 3D model of a face, and verbal descriptions of the person presented by eyewitness. Face sketch may be generated from these verbal descriptions which can be matched against mug shot database. This paper reviews various methods and techniques used to identify a person of interest in an unconstrained environment with the above-mentioned challenges.