Convolutional Color Constancy

TL;DR

This paper reformulates color constancy as a 2D spatial localization problem in log-chrominance space, applying object detection techniques to significantly improve white-balance correction performance.

Contribution

It introduces a novel formulation of color constancy as a spatial localization task, enabling the use of detection methods for better accuracy.

Findings

Nearly 40% performance improvement on benchmarks

Effective discrimination between well and poorly white-balanced images

New approach leveraging object detection techniques

Abstract

Color constancy is the problem of inferring the color of the light that illuminated a scene, usually so that the illumination color can be removed. Because this problem is underconstrained, it is often solved by modeling the statistical regularities of the colors of natural objects and illumination. In contrast, in this paper we reformulate the problem of color constancy as a 2D spatial localization task in a log-chrominance space, thereby allowing us to apply techniques from object detection and structured prediction to the color constancy problem. By directly learning how to discriminate between correctly white-balanced images and poorly white-balanced images, our model is able to improve performance on standard benchmarks by nearly 40%.