Selective Colour Restoration of Underwater Surfaces
Chau Yi Li (Queen Mary University of London, UK),* ANDREA CAVALLARO (Queen Mary University of London, UK)
The 33rd British Machine Vision Conference
Abstract
The appearance of surfaces in underwater images is degraded by the selective attenuation of light in water. The light intensity decays exponentially along the vertical depth and along the range between the surface and the camera. Images capturing a large vertical depth exhibit non-uniform water colour. Restoration methods that compensate for the resulting colour cast and reduced contrast generally ignore the degradation caused by depth, whereas methods that target the removal of the colour cast often distort the colour of the water mass. Furthermore, most methods assume a uniform water colour and under-compensate for the colour when the water colour is non-uniform. In this paper, we present a selective chromatic adaptation (SeCA) method that restores the colour appearance of underwater surfaces to that under an unattenuated light, as if the surfaces were captured in air. Using the Schechner-Karpel model, we restore the colour degraded along the range by estimating the extent of colour degradation for each colour channel. Moreover, we handle the case of uniform and non-uniform water colour with one single approach. We also derive a scene-adaptive map that restores the colour degraded along the vertical depth by selectively removing the cast on surfaces while maintaining the water colour. SeCA needs no knowledge of the range nor the vertical depth at which the surfaces are captured. SeCA outperforms state-of-the-art neural networks in terms of colour accuracy. Furthermore, we validate the stability by deploying SeCA on underwater videos without any temporal regularisation.Video
Citation
@inproceedings{Li_2022_BMVC,
author = {Chau Yi Li and ANDREA CAVALLARO},
title = {Selective Colour Restoration of Underwater Surfaces},
booktitle = {33rd British Machine Vision Conference 2022, {BMVC} 2022, London, UK, November 21-24, 2022},
publisher = {{BMVA} Press},
year = {2022},
url = {https://bmvc2022.mpi-inf.mpg.de/0228.pdf}
}
Copyright © 2022
The British Machine Vision Conference is organised by