Estimating Polynomial Coefficients to Correct Improperly White-Balanced sRGB Images

When taking photos, the illuminant of a scene can bring undesirable color casts to an image and ordinary users mainly rely on automatic white-balance to discount the effect of the illuminant. Since automatic white-balance is a nontrivial problem, many images can be wrongly white-balanced and rendere...

Full description

Saved in:
Bibliographic Details
Published in:IEEE signal processing letters 2021, Vol.28, p.1709-1713
Main Authors: Luo, Hang, Wan, Xiaoxia
Format: Article
Language:English
Subjects:
Artificial neural networks
Color
color constancy
color correction
Convolution
convolutional neural networks
Image color analysis
Kernel
Linear transformations
Mathematical analysis
Matlab
Neurons
Polynomials
Thermal expansion
Training
Transforms
White balance
White balancing
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:When taking photos, the illuminant of a scene can bring undesirable color casts to an image and ordinary users mainly rely on automatic white-balance to discount the effect of the illuminant. Since automatic white-balance is a nontrivial problem, many images can be wrongly white-balanced and rendered into standard RGB (sRGB) color space. In this paper, we propose to construct a residual image to correct wrongly white-balanced sRGB images by applying a linear transform matrix on a polynomially expanded image, in which the coefficients are inferred from wrongly white-balanced images by a deep convolutional neural network. The proposed method has been comprehensively investigated and evaluated, and experimental results have shown the superiority of the proposed method.
ISSN:1070-9908
1558-2361
DOI:10.1109/LSP.2021.3102527