Estimation of Multiple Illuminants from a Single Image of Arbitrary Known Geometry

Yang Wang and Dimitris Samaras

Computer Science Department, SUNY, Stony Brook, NY 11794-4400, USA
{yangwang, samaras} AT cs DOT sunysb DOT edu

We present a new method for the detection and estimation of multiple illuminants, using one image of any object with known geometry and Lambertian reflectance. Our method obviates the need to modify the imaged scene by inserting calibration objects of any particular geometry, relying instead on partial knowledge of the geometry of the scene.

High resolution version of some of the figures in the paper which is supplemental material accompanying our ECCV02 paper. Slides of our ECCV02 presentation.

Fig. 8(a) the original image.
Fig. 8(b) the generated image of a Lambertian ball with the five light sources extracted from Fig. 8(a).
Fig. 10(a) the 3D shape of the duck frontal surface, R,G,B color values represent the x,y,z components of the normal. Fig. 10(b) the original image. Fig. 10(c) the generated image of a Lambertian duck with the four light sources extracted from Fig. 10(b), where for completeness purposes the eyes and beak were copied from the original image manually as they have different albedo and non-Lambertian reflectance characteristics.

Notice that the recovered 3D shape in Fig. 10(a) has a high level of noise. The noise in the generated image in Fig. 10(c) is due to the inaccuracies in shape estimation. Nonetheless illuminant estimation is still possible.