A research group consisting of Professor Hiroshi Ishikawa of Waseda University Faculty of Science and Engineering, Assistant Professor of Satoshi Iizuka, and Assistant Professor of Simosera Edgar will apply artificial intelligence technology called "deep learning" to naturally color old black-and-white photographs. The method was established.It is likely to attract attention because it does not require any human intervention and can be reproduced naturally even in a scene 100 years ago.
According to Waseda University, artificial intelligence learns features that can be used as clues for coloring from a large amount of black-and-white and color images, and converts the given black-and-white images into color images by making use of those features.Artificial intelligence recognizes features obtained from the entire image and features extracted from small areas separately.From the entire screen, read whether it is outdoors or indoors, daytime or nighttime, and from small parts, guess how to color it according to the texture of each object such as sand, leaves, and water.By connecting the two and advancing coloring, it has become possible to reproduce subtle shades such as the dusk sky and human skin according to the situation.
By using this method, even black-and-white photographs taken in the Meiji era 100 years ago can be colored naturally without human intervention.When the coloring results were evaluated by user tests, it was found that about 90% of the coloring results looked the most natural.The research results will be announced at the international computer graphics conference "SIGGRAPH 7" to be held in Los Angeles, USA in July.