Current Projects

Face Recognition in Low Quality Video

Automated face recognition has taken hold in law enforcement agencies as an effective means for recognizing an unknown individual. In additional to traditional mug shot images, an immense source of face data that can be used to determine criminals identity is available from surveillance cameras. However, it is well documented that the success of face recognition in low quality images is still quite poor. In surveillance video, image frames are generally of low quality due to the compression of the signal, low resolution cameras, and unconstrained external parameters (pose, illumination, occlusion). In [1] we observed the effects of increasing levels of video compression on face recognition performance using the H.264 compression algorithm. It was determined that the video signal can be compressed up to 128kbs in most instances without a noticeable impact on face recognition performance. Future work in this direction includes an image processing algorithm that is able to enhance the quality of a video face image in order to improve its ability to be correctly matched.

B. Klare and M. Burge, “Assessment of H.264 Video Compression on Automated Face Recognition Performance in Surveillance and Mobile Video Scenarios,” submitted to SPIE Conference on Biometric Technology for Human Identification, 2010.