Artificial intelligence to restore poor quality images
Simple to use, and offering great results, such as the opportunity to enlarge images and clarify blurred areas, this new artificial intelligence system called GFP-GAN and developed by researchers at the TENCENT ARC Laboratory in China, is the latest alternative presented to restore images automatically.
The tool can be used from the web and run for free, although it has a couple of demonstration images of Abraham Lincoln and Albert Einstein, the user can upload any photograph and submit it to the restoration process.
The team presented a report specifying the technical details behind this tool, as well as algorithms used and the complexity of the generative adversarial network (GAN) architecture employed together with the generative face priority (GFP) system, where both perform face restoration through novel spatial feature transformation layers divided by channels, allowing this tool to achieve a good balance of realism and fidelity.
The restoration of facial details is performed together with color enhancement, so that satisfactory results are obtained in a single pass, in addition to the considerable reduction of resources.
Although it is a little-known alternative and perhaps designed to be used in the artistic sector, it could be of great help in places such as museums and libraries where there is a large base of old images and that, over the years, have lost quality and sharpness, in addition to being analog images, which, not being properly preserved, deteriorate more quickly.
The tool can be accessed at https://huggingface.co/spaces/akhaliq/GFPGAN and read about the report at https://arxiv.org/abs/2101.04061
