Face recognition is a comprehensive, multiple-level fusion application of multispectral image fusion. The facial images consist of two-spectral stereo images: visible and thermal (LWIR) images from left and right cameras. Fusion is conducted at four levels: image (pixel), feature, score, and decision. Face scores are computed from three algorithms: circular Gaussian filter (CGF), face pattern byte (FPB), and linear discriminant analysis (LDA). Fusion can be implemented with spectral images, stereo images, or at different levels.
Humans can recognize a face with monocular or binocular vision, whereas computers typically use only a monocular facial image. The performance of face recognition can be improved by using multispectral images and image fusion; the goal is to investigate stereo vision for a multispectral face-recognition system. Human binocular vision has many advantages, such as stereopsis (3D vision), binocular summation, and singleness of vision, including the fusion of binocular images (cyclopean image). In human visual processes, the binocular summation and singleness of vision are similar to computer image-fusion processes. A multispectral face recognition system is enhanced with stereo imaging capability in this application, which is implemented with visible and thermal cameras and three recognition algorithms (CGF, FPB, and LDA).