Face recognition, the process of automatically identifying a person from an image/video, has many important applications. For instance, in law enforcement, crime scene video footage is automatically matched to a mugshot database of previous arrests. In border control, face recognition certifies that the holder of a passport is the rightful owner by automatically matching photographs to the image stored in the e-passport microchip. Face recognition is also employed by governments to detect and deter fraud in visa and passport applications. There are two major drawbacks with current face recognition systems:

  1. A photo of the suspect must be available;
  2. Scenarios in which faces are partially occluded, disguised or showing different facial expressions cannot be handled robustly.

We propose to develop a facial-composite-based recognition system for identifying faces under partial occlusions, disguises, different facial expressions and arbitrary poses from surveillance videos and images. A facial composite is a graphical representation of an eyewitness's memory of a face. Currently, face composites are used for identifying a suspect in a wanted poster or warning the population against offenders. In this project, we propose to develop a system that enables robust face recognition using only a facial composite of a suspect.


7 May 2013
The Standard reporting on our project on 7 May 2013
The Standard on 7 May 2013
12 August 2012
SCMP reporting on our project on 12 August 2012
SCMP on 12 August 2012 SCMP on 12 August 2012
3 July 2012
The project kick off