Illumination ratio image: synthesizing and recognition with varying illuminations

摘要

Illumination ratio images are presented in the paper to illuminate the faces. Our rendering method can simulate the distribution of the images with varying illuminations and generate new training images for face recognition with single frontal view. We analyzed the reason of the assumption that different persons have the same surface normal but different albedo. The synthesized images were compared with the original images by Gamma correction and error images. Our experiment shows that eyes are not lambertian surfaces. The synthesized images improve the face recognition performance by using the individual eigenface classifier.