Multi-illuminant color constancy for HDR images through exposure segmentation

A standard practice in high dynamic range imaging is to compose an image through exposure bracketing which captures a series of exposures of the same scene and then combine them together, followed by dynamic rang compression and some color processing steps. Scenes lit by multiple illuminants such as a room with an artificial light source when the sun is shining through the window is an often encountered scenario which offers opportunity for the high dynamic range feature of an image pipeline to show its advantages. Traditional color constancy algorithms estimate a global white point of the scene and then apply color correction based on this estimate, which could exaggerate the difference between the illuminants, making part of the image better and part of the image worse, or compromise the color of the whole scene. In this paper, we propose a method for the color constancy of high dynamic range scenes with multiple illuminants utilizing the inherent difference in their luminance levels to assist the segmentation of the image into differently illuminated portions and apply their corresponding color constancy parameters. Experimental results using two exposures show superior performance of the proposed algorithm compared to traditional algorithms applying global corrections only.