How AI revives the early cinema
Dr Wan Renjie, Assistant Professor of the Department of Computer Science, was involved in the restoration and colourisation of the silent film produced over a century ago. “The quality of the film deteriorated over time, and there was a lot of noise in the visual data, which greatly affected the viewing experience. To restore the film, our team used AI technologies to remove video noise, reduce flickering and increase the resolution of the film,” he says.
Another important aspect of the restoration process was adding realistic colours to the monochrome film. To achieve this, the team trained machine learning algorithms to colourise low-quality monochrome video frames. “The techniques in image colourisation for static photos are well developed, but the challenge in colourising movies is that they contain a series of images to create a persistence of motion. Therefore, when colourising videos, we have to consider the relationship between each frame to ensure colour consistency,” says Dr Wan.
The team also employed a technique named Neural Radiance Fields that can generate 3D representations of a scene based on static images. “From the images, we can extract the necessary features for colourisation and construct a colourised video sequence,” says Dr Wan.
Professor Poon believes the technologies that powered the novel audio-visual experience at the concert can provide contemporary audiences with more opportunities to appreciate early cinema. “These technologies not only restore damaged silent films, but they also make early cinema relevant and accessible, allowing more people to better understand the cultural values and human experiences of the past,” he says.