Animals Career Life Travel

This Incredible Algorithm Can Remove the Water From Photographs

This Incredible Algorithm Can Remove the Water From Photographs July 17, 2020Leave a comment

The seaweed is always greener in somebody else's photos. Anyone who has tried capturing the beauty of the ocean has experienced the blue hue of the water in all of their photos, but recently an expert came along who taught a computer to capture underwater moments without capturing the water...

Open Your Eyes and Sea

Scientific American

Have you ever noticed how underwater photos look a bit weird? They're always very green and dull, making it difficult to really see what coral and fish actually look like in their natural habitat.

An Expert to the Rescue

Scientific American

Enter Derya Akkaynak, an oceanographer and engineer who's spent the past few years coming up with the computer algorithm to fix murky photos from below the depth of the sea.

Picture Fix

Scientific American

Akkaynak's innovative program relies on an advanced artificial intelligence that can strip away the murky, muddy color of ocean water to present pristine, accurate photos from under the sea.

Years of Work

Scientific American

To get this right, Akkaynak has been relying on a standard commercially available camera and a color chart - the logic being that if her algorithm can fix relatively inexpensive photos, it'll be of use to the most people possible.

Welcome to the Matrix

Scientific American / Derya Akkaynak

Because of the way light bounces through sea water, close-up images can end up looking washed out, with a heavy green tint. If you've ever been scuba diving you'll recognize that this isn't exactly accurate to how fish look in real life.

A Breath of Fresh Water

Scientific American / Derya Akkaynak

Akkaynak's algorithm measures the various shades of green within an image so that close-up colors can be restored, making for far more appealing images that also serve an important scientific purpose.

Limited Visibility

Scientific American / Derya Akkaynak

Another problem with sea water is that it often makes photos look muddy and cloudy. Even in relatively clear water, it's hard to capture a photo that accurately shows off objects or animals in the distance.

Color Returns to Coral

Scientific American / Derya Akkaynak

While not all of this murk can be filtered out, the algorithm can remove enough to show off the full variety of life that exists below the surface. This is good news for marine biologists who could save a lot of time thanks to these photos.

50 Shades of Green

Scientific American / Derya Akkaynak

The problem with colors that don't quite match up is that it's hard to identify what marine plant or animal you might be looking at. Akkaynak's color chart shows just how inaccurate underwater photos can become.

Speeding Things Up

Scientific American / Derya Akkaynak

With no clear photos to work from, biologists studying specific animals have a difficult time working from photographs. Often they'll have to identify fish in person or scour images personally, wasting a lot of effort and resources.

Robots are the Future

Scientific American / Derya Akkaynak

With better, clearer photos, these biologists can analyze images with greater ease to spot what they're looking for. Additional AI algorithms could even be trained to search pictures on a scientist's behalf.

State of the Art

Scientific American / Derya Akkaynak

The algorithm, called Sea-Thru (because of course it is) isn't just a Photoshop filter. It's not enough to adjust a setting and suddenly see all the vibrant color of marine life spring back into existence.

Crystal Clear

Scientific American / Derya Akkaynak

This is why creating Sea-Thru has taken years and years of work. It's important that the algorithm can tell, simply by looking at an imagine, exactly what color it would be if viewed outside of the water.

Looks a Bit Fishy

Scientific American / Derya Akkaynak

This process has involved photographing reefs from a variety of different angles, giving the computer as much data as possible to work from in learning real oceanic colors.

All Angles

Scientific American / Derya Akkaynak

According to Akkaynak: "Every time I see a reef with large 3D structures, I place my color chart at the base of the reef, and I swim away about 15 meters. Then I start swimming towards the reef, towards the color chart, and photograph it from slightly different angles until I get to the reef."

Artificial Intelligence

Scientific American / Derya Akkaynak

A lot of people worry that artificial intelligence will take away their jobs over the next few decades. A lot of jobs will certainly change, but AI will be most useful as a tool to help humans' lives become more efficient.

Smart Machines

Scientific American / Derya Akkaynak

Sea-Thru is just one of many AI algorithms that are making an impact in the scientific world. Computers can even identify potential tumors on patient x-rays with a surprising degree of accuracy.

Pretty as a Picture

Scientific American / Derya Akkaynak

While Sea-Thru will be useful for science, it will also make a big difference to commercial photographs of aquatic life. Get ready for a future where nature photographers and undersea documentaries can better show off the vibrant ocean world. At least for a while.

Depressingly Accurate

Scientific American / Derya Akkaynak

Another benefit of Sea-Thru is that it'll make monitoring changes to sea habitats a lot easier. This will help those scientists who are studying the effects of the ongoing man-made climate crisis on reefs that are currently dying.

A Final Archive

Scientific American / Derya Akkaynak

Considering the rate with which reefs are dying (it's happening very rapidly), this algorithm will also help to archive images of many aquatic creatures who are about to disappear forever.

Lots of Work

Scientific American / Derya Akkaynak

In order to train her algorithm, Akkaynak had to take well over a thousand photos, all of which contained her color chart to give Sea-Thru a base guide for what the photo should look like.

Accessories Not Included

Scientific American / Derya Akkaynak

Now that the algorithm has been properly trained, however, the color chart is no longer necessary. All it needs is a few photos snapped under normal lighting conditions, and it can work out accurate colors for itself.

A Real Timesaver

Scientific American / Derya Akkaynak

Akkaynak anticipates that the algorithm will be particularly useful to any scientist who is trying to color correct thousands of images at a time, as it can be set to work quietly in the background without any need for human intervention.

Pooling Data

Scientific American / Derya Akkaynak

This automation will remove one of the biggest drains on researchers' time, especially for those who are working with large data sets that need to be analyzed— even clicking once or twice on each photo to optimize them would take hours and hours if done by human hand.

Better Down Where It's Wetter

Scientific American

With any luck, Akkaynak's work will make a big difference to marine biology going forward. If we also get some pretty photos of life under the sea, all the better!