Berkeley image colorization presentation scheduled

August 1, 2017 by  

Researchers from the University of California, Berkeley have presented their work on data-driven image colorization at SIGGRAPH 2017.

The innovation in image colorization developed at Berkeley is data-driven and employs artificial intelligence. The technique works in real-time, enabling even novices to take grayscale images and colorize them with pleasing results. A Convolutional Neural Network (CNN) takes user edits in combination with information gleaned from large-scale data it has been provided.

As the CNN learns what colors are common for various objects, it is able to make appropriate suggestions to the user. However, the system, although trained to know a cow’s natural color options, will oblige with a purple cow if the user wishes. This is a step up from preceding automatic colorization systems that did not allow real time color adjustments.

CNN testing proved itself to be user friendly. Novice users, provided with minimal training, were able to produce colorization undetectable from real images. It is likely only a matter of time before this technology is adapted for use by printing companies.

SIGGRAPH is the biggest, most prominent annual conference for computer graphics in the world. For five days, attendees review research results, attend educational sessions, view demos and witness hands-on, cutting-edge, interactive features. Three days of the conference included commercial exhibits showing off the industry’s latest applications, including the available hardware, software, and services.

This year’s event is being held at the Los Angeles Convention Center from July 30 through August 3.