I just stumbled upon Scientific American news of See Thru, and had searched the original paper. The result looks great for removing color cast. Is the original pdf of use? I’ll attach it since I cant link it from my phone.
Goes with this thread:
Interesting, I’ll have to take a look at this tonight.
I would not be surprised if this is a variation on the existing “Dehaze” algorithm found in many software applications and could be implemented as an alternative mode of that algorithm.
It doesn’t seem possible according to @afre in the linked thread. However, I don’t think that this thing shouldn’t be done at all, it may be possible that it’ll become useful in the future.
My point on the other thread is to temper your expectations. The paper isn’t some magical incantation. You need the correct input data for the algorithm to run at all, and won’t be able to do that with photography as we know it. We need specialized equipment under ideal experimental conditions.
My mini-essay is meant to break down the process into understandable chunks. With existing tools, there is a lot that we can do.
I just found a paper about a way to extract a focus map from a single picture (with no deep-learning). It’s not exactly the depth map that this algo expects, but it’s the closest I have found so far:
An In-Depth Survey of Underwater Image Enhancement and Restoration (PDF). Have fun!