Make black and white photos blush again with this AI-powered tool

Do you have a black and white photograph you just wished was in full colour? Well, this clever neural network tool from Waseda University academics might be just what you’re looking for.

The tool uses machine learning to intelligently (or, as intelligently as possible) shade a black and while image with appropriate hues, shades and tints. It effectively re-colours a black and while photograph based on features found within it.

For instance, golf courses, hills and trees will naturally contain greens, while abbeys, arches and statues likely won’t. The tool takes this information, and through best learned guess, adds colour.

Using machine learning and a neural network, the tool can accurately add colour to black and white images

The team based in Tokyo consisted of Satoshi Iizuka, Edgar Simo-Serra and Hiroshi Ishikawa.

“Based on Convolutional Neural Networks, our deep network features a fusion layer that allows us to elegantly merge local information dependent on small image patches with global priors computed using the entire image”, explains the team’s abstract.

The system uses four effective layers to decode a photograph. A “low-level features network, a mid-level features network, a global features network, and a colorization network” work together to output an image with vibrant, and fairly accurate colour.

According to the team, the technology can process images of “any resolution”. That means if you have a relatively small photograph, or a much larger RAW image, it should all be compatible.

As for its accuracy? Well, have a look at the video below.

This isn’t the first time we’ve seen neural networks used in image colourisation. Colourzebot came to internet fame back in July 2016, but wasn’t absolutely accurate. Still, people definitely had fun with it.

If you’re interested in giving the tool a go, navigate to this web page, drag your image into the box (or select the Browse button), and hit “Colorize!”

Feature image: Waseda University/Satoshi Iizuka, Edgar Simo-Serra and Hiroshi Ishikawa