3-25 A New Face-Recognition System with Robustness against Illumination Changes
Rui Ishiyama, Shizuo Sakamoto, Johji Tajima · 2000
Proposed here is a new face-recognition system based on 2D / 3D scheme. It acquires an ordinary 2D image as a query, which is then compared with each registered 3D shape and surface reflectance. The authors have developed two technologies: a novel range finder, and an image compensation technique to cope with changes in illumination. The range finder employs a new design: sinusoidal grating projection combined with stereop-It offers accurate measurement (0.2mm, rms) of all pixels in a measuring time that is short and if such changes outweigh individuality, recognition becomes difficult [I]. The three dimensional nature of the face makes this problem especially complex. If the 3D shape of a face were registered in a face recognition system, it shoul, however, be possible to estimate and compensate for image changes sufficiently to recognize individuals with significant accuracy. We are developing a three dimensional face recognition system in which the input is a 2D image taken by an ordinary camera. The reference data, however, is based on a 3D shape and the (2 set) be for a number associated surface reflectance data. of human applications. The 3D shape enables us to compensate for image changes due to changes in illumination. But it is difficult to parametrize illumination because the number of parameters which describe illumination is arbitrary due to the number of light sources. Here the authors reveal that the illumination subspace constructed by quite a few basis images cover almost all the illumination changes. We conducted recognition experiments using images of 42 people in four poses and under four drastically differing sets of illumination conditions, for a total of 672 query images. Our method achieved a first-choice success ratio of 97 percent. 1