Illuminant estimation method based on Color Lines and dichroic reflection model

Abstract

In some shooting environments, a color cast can be introduced by the color of the illuminant. This color cast can result in an image that differs from the original color of the subject, adversely affecting image analysis and recognition processes that rely on accurate color information. White balancing is a technique to eliminate the effects of illuminant and the associated color cast. The effectiveness of white balancing depends on an accurate estimation of the illuminant. Various methods for illuminant estimation have been proposed, including hypothesis based approaches, deep learning methods, and methods based on the dichroic reflection model. However, these methods are insufficient to remove the color cast for the image with a distorted color distribution and/or noise present. In this paper, we propose an illuminant estimation method using Color Lines in conjunction with the dichroic reflection model. In the proposed method, first, specular reflection is calculated based on the dichroic reflection model using a thresholding process to eliminate halation. Subsequently, clustering is applied to the calculated specular reflectance to segment the image into regions affected and unaffected by the illuminant. Meanwhile, it has been reported that Color Lines, which represent the color distribution within local regions of the same object as straight lines, intersect near the color of the illuminant. These intersections are used to identify the region most affected by the illuminant from the clustered specular reflections, which is then estimated as the final illuminant. In the experiments, the effectiveness of the proposed method is verified through both subjective and quantitative comparisons with conventional methods.