Re-lighting and compensation for face images

Abstract

The illumination variance makes the robust face detection a challenging problem. Although there are some methods focusing on the de-lighting problem, the requirement for testing set or the 3D information limits the application for the detection tasks. According to the reflection function, the illumination takes the role of amplification for the reflective character of the surface. The shadows will make the features of the object attenuated or diminished. We introduce the radiance map ratio to adjust the image to a new illumination condition and apply the original image to compensate the adjusted image. The re-lighting and compensation makes the illumination condition uniform and keeps the original smooth changed information. As a pre-filter in the detector, the re-lighting and compensation method facilitates the performance of the detector. © Springer-Verlag Berlin Heidelberg 2005.