Extracting intrinsic images from multi-spectral

Abstract

Intrinsic images as a useful midlevel description attract more and more attentions in computer vision. According to Barrow and Tenenbaum's theory, a face image can be decomposed into two images: a reflectance image and an illumination image. Finding such decomposition remains difficult since it is an ill-posed problem. In this paper, we focus on a slightly easier problem: given a pair of multi-spectral facial images, can we recover its reflectance image and corresponding illumination image? Experiments show that it is promising and feasible. According to recent research in skin color model and Quotient Image, we propose a simple but effect method to derive the intrinsic image from a near infrared and a visual image. After modulating the grey distribution of visual images and dividing visual images by near infrared ones, we can recover its reflectance and illumination image. Experimental results show that our method is promising in image synthesis and processing.