Re: Algorithm that removes the water from underwater images
PHXAZCRAIG wrote:
I like the idea, which seems aimed for accuracy as much as anything. But also in shallow water perhaps, where there is still a range of colors to be had.
Yes. This is what it does. Goal is accuracy not perse a pleasing picture. Any photo-shopping comes down to what you think it would/should look like.
Quote from developer/scientist in yet another discussion:
But I believe you (and not you personally, but those who say "I can do this in Photoshop") are still missing the point about the method. I will try again to explain below, but please don't hesitate to ask if anything is still unclear. I know underwater photographers are very passionate, and I want to make sure we all understand each other.
1) You can of course do ANYthing in Photoshop - as one person commented, you can even paint the whole image pixel by pixel with the colors you want. But any correction you do in Photoshop is subjective. Give the same raw underwater image to 10 photographers, and you'll end up with 10 different corrections. Not to mention tedious. Can you imagine doing a survey of a reef, coming back with 1000s of photographs, and trying to correct them manually one by one in Photoshop ? (Actually that's how most marine scientists work at the moment.....)
Sea-thru operates on a physical equation describing of how light propagates in the water, it knows which parameters govern how light is absorbed and scattered, and it calculates those parameters for EACH PIXEL, and makes sure each pixel satisfies that equation. Because it always makes sure all pixels obey that equation, its corrections are objective, and repeatable. And it is automated.
2) Yes, in Photoshop, you may achieve a visually pleasing correction that will look great for many images. For the most part, these images will be the ones where the scene is close to the camera (low backscatter), and more or less all objects in the scene will have the same range from the camera (uniform attenuation). [Correcting scenes with varying in Photoshop ranges will get very difficult or very tedious. And yes, your comment about examples not showing that case is valid.]
Even so, the correction you end up with will likely not be physically accurate correction because (at this point), photoshop doesn't give you the tools to extract/estimate the parameters you need to do a proper correction. Until last year, what those relevant parameters are were not even fully understood, please have a look here:
http://openaccess.thecvf.co...
There is currently a "dehaze" filter in photoshop which stems from research, just like mine, from a field called atmospheric dehazing. But that filter won't work for underwater images because the physics of light propagation in the atmosphere and in the ocean are different.
So you can imaging Sea-thru will be a filter in Photoshop just like "dehaze", soon.
All of this might sound like overkill for non-scientific underwater photographers, amateur or professional. But this work comes from a scientific field called "underwater computer vision", and is intended to accelerate our pace of progress in marine science, by standardizing large sets of underwater images, so that we can use powerful AI methods on large datasets. As a bonus, it will be a cool tool for non-scientific underwater photographers too.
Hope this helps.
Derya