Abstract
The recognition of thermal face is a very promising strategy in biometrics. It is invariant to illumination, robust to pose and immune to forgery. However, thermal face image consist of face heat energy and face counter information mainly, and it makes lower discrimination for inter-class. In this paper, an enhanced thermal face recognition approach based on Multiscale Complex Fusion for Gabor coefficients (MCFG) is proposed. Initially, the Complex Gabor Jet Descriptor (CGJD) is acquired based on the block mean and standard deviation generated from the magnitude, phase, real and imaginary parts of Gabor coefficients. Then, the Complex LDA (CLDA) algorithm and feature level fusion are implemented on multiscale Gabor coefficients to reduce the dimension and enhance the discrimination. Experiments conducted on two thermal face databases NVIE and IRIS, which have some challenging thermal face images, show that the proposed thermal face recognition approach significantly outperforms the state-of-the-art approaches.