Researchers at Waseda University in Tokyo have created a way to realistically colorize black-and-white photos without any human intervention for the first time ever. The team's approach is based on convolutional neural networks — a type of machine learning originally inspired by the visual cortex of a cat.
The researchers used artificial intelligence to classify a full image and then identify parts of that image to label its components before filling them in with the appropriate colors. Previous research efforts in automated colorization fell short of being totally automatic. Most required users to provide a reference image that was similar to the black-and-white image in order to colorize it properly. One recently proposed approach was able to fully colorize images, but it was limited to certain categories of photos like outdoor scenes.
The Waseda team showed their approach was able to get the job done much more effectively than other leading methods, and went about colorizing a whole bunch of images to show what their program could do.