Illumination estimation based on a weighted color distribution

Abstract

Based on the observation that whitish surfaces are perhaps a better indicator of the color of the illuminant than strongly colored surfaces, we propose a new illumination estimation method that attributes varying levels of significance to the whitish surfaces according to how strongly colored they are. A Gaussian likelihood function centered at the chromaticity of the training illuminant is used as the weighting function. The method is tested on both real and synthesized image data. The results indicate that the proposed method outperforms the closely related color constancy methods.