Recovering Intrinsic Images from Weighted Edge Maps

Abstract

Intrinsic images refer to the reflectance image and the illumination image. By separating the reflectance image and the illumination image from an observed image, the effects of shadows and variable illuminations on processes such as image segmentation and object recognition are considerable attenuated. An observed image is a product of its reflectance image and illumination image. We present an intrinsic image decomposition method for a single color image by assuming that sharp chromaticity changes results from reflectance changes. We first compute gradients on chromaticity. Then we multiply the edge maps of color intensity image of original image with the chromatic gradients and obtain weighted edge maps. Through weighted edge maps, we can distinguish the edges from reflectance and those from illumination. Finally, we reintegrate reflectance edge image and illumination edge image to recover the reflectance image and illumination image. The experimental results on real data are given.