Near infrared face recognition using orientation-based face patterns

An orientation-based face recognition method is proposed and applied to near infrared images in this paper. Two algorithms, FPW (face pattern word) and FPB (face pattern byte), are derived from the orientation-based method. The FPW (or FPB) algorithm actually extracts the orientational information using a set of Gabor wavelet transforms, and uses Hamming distance (HD) for face identification. The bit code of orientations are optimized with respect to HD and put into a 32-bit word (FPW) or 8-bit byte (FPB). The performance is evaluated by face recognition rate and compared with three classical algorithms, PCA, LDA and EBGM (elastic bunch graph matching). Our experiments are conducted with a ASUNIR face database that currently consists of near infrared (NIR) images from 79 subjects. The experimental results show that the FPW algorithm achieves 98.43% of recognition rate; while the results of PCA, LDA and EBGM are 74.68%, 94.94% and 96.20%, respectively.