Color correction for multi-camera systems: aligning white points

Ruikai He · Minchen Wei · Xinchao Qu · Tao Hu

🔖 Abstract

In recent years, many devices (e.g., smartphones and drones) contain multiple cameras, allowing an easy change of different focal lengths and fields of view (FOVs) when capturing ascene. Obvious color differences are easily observed when the images are captured by the different cameras for the same scene with different FOVs in asequential order, which significantly affects the user experience. Most studies and investigations attributedsuch color differences to the different spectral sensitivity functions of the cameras. In this study, we found that the differences could be mainly due to the inconsistency among the white points estimated from the images captured by the different cameras. Based on this finding, we propose a method to calibrate the cameras under a set of calibration light sources that are carefully selected and to use a conversion matrix to align the white points among the images in real-time, which was found to significantly improve the color consistency.

🔍 Introduction

A demo code for the journal paper "Color correction for multi-camera systems: aligning white points".

The demo results differs from those shown in the journal papers due to the variation of the SPDs involved.

In demo, the publicly available SPDs dataset[1], instead of the laboratory SPD collection, are utilized to select a set of representative SPDs for aligning white point.

Apart from SPD dataset, the QEs of main and telephoto camera embed in drone are not disclosed yet. In demo, the data calculated by QEs were precalculated and saved as .mat files, where suggests line 91-92 be replaced with line 94-95 in mainFunc_visualization.m.

mainFunc.m and mainFunc_visualization.m are the main functions to get the results and visual comparison, respectively. It is northwhile observing that visualizations are processed by white balance and gamma correction without color space conversion.

💻 Quickstart

Prequisition

Matlab 2018b or higher (finished on Matlab 2018b, and tested on Matlab 2023b)

Get started

1. RUN mainFunc.m for the set of selective SPDs.
2. RUN mainFunc_visualization.m for visual comparison.

Contents

1. the folder camera_spectral includes the QE (Camera Spectral Sensitivity Curves) from "What is the space of spectral sensitivity functions for digital color cameras?"[2]. All .mat files are sorted out by Yuteng Zhu's Phd Disseration and related work[3].
2. the folder data includeS the data necessary.
   1. illum_test.txt and illum_train.txt are the publicly available SPDs dataset[1].
   2. WPsTrain1x.mat and WPsTrain3x.mat are calculated by $SPD \cdot QE$.
   3. Results.mat is the selected SPDs for different camera pairs.
3. the folder imgsIllustration includes one set of images, 1x and 3x means that captured by the main camera and telephoto camera, respectively. They recorded the same scene but in different FOVs (Field of View). .mat files are the demosicked ones corresponding to .DNG images.
4. the folder internal stores the necessary sub-functions. cal_VoreValue.m is from Zhu's Phd work[3].
5. ⭐🌟mainFunc.m generates ./data/Results.mat but not the ultimate results. The ultimate result need to be processed to get indices SPDLab(:,IllumFreq(:)): the ultimate results (several representative SPDs) at the end of mainFunc.m (using the following code cell to get several representative SPDs):

clear all
% all data obtained in loop
data = load('./data/Results.mat');
ArrayIllum = data.ArrayIllum;
SPDLab = data.SPDLab;
tableSave = data.tableSave;
LisInputInter = data.LisInputInter;
clear data

for i = 1:size(ArrayIllum,1) % numIntervation
    for j = 1:size(ArrayIllum,2) % every input (R/G, B/G)
        ArrayIllumTemp = ArrayIllum(i,j,:); 
        ArrayIllumTemp = ArrayIllumTemp(:);
        [unique_elements, ~, idx] = unique(ArrayIllumTemp);
        element_counts = accumarray(idx, 1);
        [~, idxMax] = max(element_counts);
        IllumFreq(i,j) = unique_elements(idxMax);
    end
end

📧 Citation

If you find our work helpful, please cite our paper.

@article{He:25,
author = {Ruikai He and Minchen Wei and Xinchao Qu and Tao Hu},
journal = {J. Opt. Soc. Am. A},
keywords = {Cameras; Color difference; Imaging systems; Neural networks; Real time imaging; Spectral discrimination},
number = {5},
pages = {690--698},
publisher = {Optica Publishing Group},
title = {Color correction for multi-camera systems: aligning white points},
volume = {42},
month = {May},
year = {2025},
url = {https://opg.optica.org/josaa/abstract.cfm?URI=josaa-42-5-690},
doi = {10.1364/JOSAA.550307},
}

📖 Reference

[1] K. Barnard, L. Martin, B. Funt, and Adam et al., "A Data Set for Colour Research,", Color Res Appl., 27(3), 147-151, (2002).
[2] J. Jiang, D. Liu, J. Gu and S. Süsstrunk, "What is the space of spectral sensitivity functions for digital color cameras?," in 2013 IEEE Workshop on Applications of Computer Vision (WACV), Clearwater Beach, FL, USA, (2013), pp. 168-179.
[3] Y. Zhu, "Designing a Colour Filter for Making Cameras more Colorimetric," EPrint 2021.12 (School of computing science, University of East Anglia, 2021).