Infrared-based Face Identification System Via Multi-Thermal Moment Invariants Distribution

This paper presents the implementation of moment invariants in an infrared-based face identification system to develop a novel and less complicated face identification system in thermal spectrum. Initially the system decomposes an infrared image into 4 thermal regions via 3-valued threshold method. In order to increase the information obtained from each decomposed IR image, the first four of the classical Hu's moment invariants were computed from each generated thermal regions. Additional to this, centroid point and area (in pixel) were also obtained from each generated thermal region. The acquired Hu's moment invariants together with the acquired centroid point and area (in pixel) forms a string of features for classification/identification. This system employs a hierarchical minimum distance measurement method for classification purposes. The performance of this system is encouraging with 87% of correct identification rate for test to registered image ratio of 2:1 and 84% of correct identification rate for test to registered image ratio of 4:1.