Many old films and documentaries shot in black and white have been restored to give them color and make them look more current and interesting to the audience. Also, this procedure involves several different jobs. You have to increase the resolution, eliminate the noise, adjust the contrast and replace those damaged parts.
However, although innovations had been presented to automate the work, this process was often done by hand and cost a lot of work. Despite this, the results were not always of the expected quality.
Now two researchers have published a paper in which they propose to use artificial intelligence for this tedious task, saving time and with a system capable of carrying out a complete remastering of the works and that you can learn as you go.
“Our work is based on temporal convolutional neural networks with video-trained attention mechanisms with data-based impairment simulation. Our reference source allows the model to handle an arbitrary number of reference color images to color long videos without the need for segmentation, while maintaining temporal consistency, ”explain Satoshi Iizuka and Edgar Simo-Serra, authors of the innovation.
First, a sequence of black and white images is input that would be restored using a processing network and used as the luminance channel of the final video output. In a second step is when the reference color images are applied and the technology ‘tints’ these images of the films producing the final chrominance channels of the video.
More footage, better results
One of the peculiarities of the technology devised by the two researchers is that it becomes more accurate and precise the more footage it treats . Thus, with longer movies and videos, the tool improves its performance, unlike what happened with traditional systems. This also happens the higher the number of reference color images assigned to it. The more references, the better the final results.
As can be seen in a video, the results are quite incredible. With only 2 reference images, the solution of the two researchers would be capable of restoring a thousand frames of any audiovisual work.
To test the effectiveness of their solution, the duo of researchers compared the results with another approach presented by researchers Zhang and Vondrick, who also tried to color vintage movies using artificial intelligence a couple of years ago.
The innovation will undoubtedly appeal to many production companies and film studios that have restored classic films with more pain than glory. At the moment Iizuka and Simo-Serra have not revealed when it would be ready for commercial application and what the cost of its use would be.
Artificial intelligence has been shown in recent years as a great ally for the world of cinema through CGI images , which can recreate actors who have passed away using photos or videos that these interpreters starred in the past.
