BMVC 2004, Kingston, 7th-9th Sept, 2004

A Unified Framework for Scene Illuminant Estimation
W. Zhou and C. Kambhamettu (University of Delaware, USA)

Most illuminant estimation algorithms work with the assumption of one
specific type of the light source (e.g. point light source or directional light
source). This assumption brings up two main limitations which significantly
restrict the applicability of the algorithms: First, the knowledge about the
type of the light source presented in the scene is needed a priori; second, it
can not handle complex scenes where multiple different types of light sources
co-exist. To overcome these limitations, we remove the assumption about the
source type and develop a general light source model for all different types
of light sources. Based on this general light source model, we propose a uni-
fied framework to estimate multiple illuminants of different types. Within
the framework, we use an experiment setup where a calibration sphere with
a specular surface is utilized to probe the scene illuminants and a novel ray
tracing and matching algorithm is devised to estimate the light source parameters.
Experiment results demonstrate the accuracy of our method on a
variety of real images.
(pdf article)