The imaging issue in an automatic face/disguise detection system

Abstract

Automatic face recognition systems have made great strides. They still, however cannot cope with changes due to lighting and cannot detect disguises, Both of these issues are critical for the employment of face recognition systems in high security applications, such as embassy perimeters, federal plazas, mid the like. We propose novel imaging solutions that address these difficult problems. We demonstrate with theoretical and experimental arguments that a dual-band fusion system in the near infrared can segment human faces much more accurately than traditional visible band face detection systems. Face detection is useful by itself as an early warning method in certain surveillance applications. Accurate face delineation can also improve the performance of face recognition systems in certain difficult scenarios, particularly in outside environments. We also demonstrate with theoretical and experimental arguments that the upper band of the near infrared (1.4-2.4 /spl mu/m) is particularly advantageous for disguise detection purposes. This is attributable to the unique and universal properties of the human skin in this sub-band. We conclude the paper with a description of our ongoing and future efforts.