Underwater images suffer from attenuation and scattering of light, making them appear blueish/greenish and foggy. We want to remove such nuisances from images, but at the same time also find out, what are e.g. the water parameters or how the light source looked. Both problems are tighly coupled.

Since underwater light transport is quite complex in general, it is difficult to solve for water parameters or light source parameters using closed form equations. However, by treating computer vision as an inverse rendering problem, we can use physically correct volumetric raytracing to identify parameters from images. Mathematically, this uses equivalent frameworks like training neural networks.

See also:

D.Nakath, M.She, Y. Song, K.Köser:
In-Situ Joint Light and Medium Estimation for Underwater Color Restoration.
In: Computer Vision in the Ocean WS, International Conference on Computer Vision. IEEE, 2021. doi: 10.1109/ICCVW54120.2021.00416
[Open Access Version of Article provided by the Computer Vision Foundation]