Color Constancy: How to Deal with Camera Bias?

Yi-Tun Lin (University of East Anglia),* Bianjiang Yang (Purdue University), Hao Xie (Meta Platforms, Inc.), Wenbin Wang (Meta), Honghong Peng (Meta), JUN HU (Apple Inc)
The 34th British Machine Vision Conference


In color constancy, we seek to estimate and remove the color of the illuminating light from the captured raw RGB images. While the learning-based methods perform better than the traditional statistics-based methods, they are typically tuned to perform well on a particular camera, i.e., their performance on images captured by other cameras is subject to their cross-camera generalizability. This problem was partially addressed, in various prior works, using deep learning architectures. In this paper, we examine how these cross-camera models perform compared to where we “pre-calibrate” the camera biases in the training and/or testing images using a simple homographic color correction procedure which can be easily done on the camera manufacturer’s part. And further, with pre-calibrated data we examine by how much we could simplify the original cross-camera models. We show that cross-camera color constancy with a simple pre-calibration process yields up to 36% performance boost, which in turn indicates that the original cross-camera methods have only limited ability to compensate for the cross-camera prediction bias. Surprisingly, with this newly proposed evaluation protocol, we also found that some single-camera color constancy algorithms already possess cross-camera ability similar to adopting a camera bias pre-calibration.



author    = {Yi-Tun Lin and Bianjiang Yang and Hao Xie and Wenbin Wang and Honghong Peng and JUN HU},
title     = {Color Constancy: How to Deal with Camera Bias?},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year      = {2023},
url       = {}

Copyright © 2023 The British Machine Vision Association and Society for Pattern Recognition
The British Machine Vision Conference is organised by The British Machine Vision Association and Society for Pattern Recognition. The Association is a Company limited by guarantee, No.2543446, and a non-profit-making body, registered in England and Wales as Charity No.1002307 (Registered Office: Dept. of Computer Science, Durham University, South Road, Durham, DH1 3LE, UK).

Imprint | Data Protection