Towards the Estimation of Non-Uniform Illumination in Real-World Scenes

In a raw photographic image, the observed colors are a mixture of material reflectance and colored illumination. For further processing, the color influence of the light source is often considered a disturbance. In such cases, it is desireable to neutralize the color of the light source, often referred to as white balancing. Successful white balancing often requires an estimate of the color of the light source. Most existing illuminant color estimators assume homogeneous, globally uniform illumination. However, this assumption is often violated in real-world images. In this work, we make an attempt towards estimating the colors and spatial distributions of multiple light sources in a scene. We estimate the color of the illumination locally, and define an objective function to select two or three representative illuminants from all local estimates. Although the objective function is relatively simple, qualitative results look promising.