Coloring the monochrome days: a new technique for colorization of grayscale images
A research group led by Professor Hiroshi Ishikawa and Junior Researchers Satoshi Iizuka and Edgar Simo-Serra of the Faculty of Science and Engineering at Waseda University established a technique for automatically colorizing grayscale images by using convolutional neural networks, a type of deep learning by artificial intelligence.
Colorized photograph of Shigenobu Okuma and guests, the Okuma Greenhouse, 1910s
Previous American presidential candidate Brian in the center, the Okuma Greenhouse, Oct. 18, 1905
Shigenobu Okuma, Taisho period
The old library (now Building 2), 1933
Inside the Okuma Greenhouse, the end of the Meiji period
The colorization architecture of this model is divided into four parts: a low-level features network, a mid-level features network, a global features network, and a colorization network. This colorization method differs from previous ones for it is completely automatic and does not require any reference images.
This technique, partially supported by Japan Science and Technology CREST Program, will allow natural coloring of different kinds of images, including monochrome photographs dating back 100 years. The results of this research will be presented at an international computer graphics convention, SIGGRAPH 2016, in Los Angeles in July.
The source code is available on GitHub: https://github.com/satoshiiizuka/siggraph2016_colorization
Images from the Waseda University Archives