Detecting Multiple Illuminants

The presence of multiple illuminants in an image is an obstacle for many computer vision algorithms as well as most illuminant estimation algorithms. Multiply lit scenes do, however, occur in many situations, such as the presence of shadows or daytime indoor environments. Common ways to minimize the problems created by multiple lights are to perform a given task only in regions containing a single light or to remove the non-prevailing illumination, thereby rendering the entire image under a single illuminant. Before these methods can be implemented though, it is necessary to first detect the parts of the image that are differently illuminated. In this paper, we propose a novel method to detect multiple illuminants using the chromagenic theory. We start by taking two pictures of each scene, a normal one and one where a colored filter is placed in front of the camera, and then estimate the illuminant incident to each pixel of the