Sea-Thru: A Method for Removing Water from Underwater Images

&quot;I really see this as a start of the artificial intelligence boom in marine science&quot; - first words of the video in the article.<p>I don&#x27;t understand why to even say something like that. PR value? As I understood the paper, the presented method is very refreshingly clean of anything resembling what we call AI today. It seems to be a combination of good, old-school photogrammetry and image processing techniques - which is great, because with such methods they can actually ensure the result is physically correct.

Neat! This uses Structure from Motion to compute depth to each pixel and correct backscatter for the specific distance light is traveling from the subject. Typical flat color correction algorithms can help reduce the blue tint typical of underwater photography, but it&#x27;s only physically correct for a narrow distance band.

A couple of days before I saw the Scientific American article someone dropped this project in our local development group: <a href="https:&#x2F;&#x2F;github.com&#x2F;nikolajbech&#x2F;underwater-image-color-correction" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;nikolajbech&#x2F;underwater-image-color-correc...</a> the repository contains some test images, and they also provide a website to test it out: <a href="https:&#x2F;&#x2F;colorcorrection.firebaseapp.com&#x2F;" rel="nofollow">https:&#x2F;&#x2F;colorcorrection.firebaseapp.com&#x2F;</a> that states: &quot;This web-application features a beta version of the Paralenz color correction algorithm, which features an open-source license&quot;.<p>The two projects seems unrelated at first glance, but the timing is interesting.

I&#x27;d love a tool to reverse this process on normal photos so I can put random family events under water without drowning the family

<a href="https:&#x2F;&#x2F;www.scientificamerican.com&#x2F;article&#x2F;sea-thru-brings-clarity-to-underwater-photos1&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.scientificamerican.com&#x2F;article&#x2F;sea-thru-brings-c...</a><p>Original article linked in current article, has more pictures&#x2F;content.

While, as explained in the article, the method was developed to help scientists and not primarily to improve aesthetic qualities, I find most of the images very pleasing.<p>The one thing that is disturbing though is images with visible <i>&quot;horizon&quot;</i> (for lack of a better word). I find the images that look like they were taken on land but don&#x27;t have a sky where one would expect it somewhat uncanny.

These are truly amazing results. Very excited to see some wonderful underwater photography in the coming years. Also, the name &quot;Sea-Thru&quot; is cleverly perfect!

I would be interested in an experiment of exporting video to a frame sequence, and then letting the software process each frame. Would it make the same decisions on each frame so that when played back as video there is no visible changes between frames? While the results are impressive for a single still, using this to see movement that occurs underwater would be amazing with the improved color.

I kinda want to see this applied to the diver video. It could surface artifacts, or look surreal.

I was initially disappointed by the shadows&#x2F;refraction patterns of the surface waves are present in the images. Can we compensate for that?

This seems like it&#x27;s just color fixing using stereographic distance modeling and a known reference palette. I&#x27;m not sure what is new or novel about this? Maybe it&#x27;s just new to oceanographic photography?

is it physically accurate image though? I think those untouched photos are accurate same as seen by naked eye<p>sure it&#x27;s more aesthetically pleasing, but it&#x27;s already distorting reality<p>personally I think photos should be most accurate representation of what healthy naked human eye see, no beatification, no bigger contrast or oversaturating despite making photos more appealing. if you twist reality where it will stop, where is the border of what is too much?

I&#x27;m not really excited about technology anymore. But I found this absolutely amazing. The people that made the jankey whale pop out of the floor and raised billions have nothing on this. I&#x27;m not joking about how cool I think this is.

Does something like this exist for over-water photography? I remember once taking a picture of beautiful scenery in the Japanese mountains, and the result was a very disappointing, almost completely grayed out image.

The name is pretty good.

How does one verify if the output is correct?

The branding of this is on point. Well done.

Whether the modified photos look better or not is mostly a matter of taste I guess. Personally I prefer the non-modified photos for the underwater feeling they give. It&#x27;s a cool technique though, and probably useful for scientific purposes.

r_wateralpha 0.1