Photo: Morello Gustinelli/Reproduction
Researchers at Tencent ARC Lab – located in China – have developed a new artificial intelligence (AI) system that allows you to highlight faces in old, damaged or faded photos. The problem is that the new algorithm has some “side effects”.
The idea is good. Dubbed GFP-GAN – short for “Generative Face Priority” by its “Generative Adversarial Network” architecture – the system works like any other tool that seeks to improve images using artificial intelligence.
The difference with the Chinese algorithm is that it intends to be more accurate than previous AI systems, by focusing on honest facial details to preserve facial identity, and striving to strike a balance between reality and sincerity.
In the tests carried out, the system achieved excellent results in very blurry images, as in the photo below.
However, when using high-resolution images, the app ended up creating rather scary images.
Far from the innocent benefit of improving older images, the use of systems like GFP-GAN could raise privacy concerns.
Artificial intelligence like this can be used to monitor populations, and improve CCTV images of people who are far away or of poor quality to identify and track specific individuals.
Today, surveillance is already used by businesses and governments around the world, but it has the limitation of not providing clarity to images of people and distant objects.
Improving this type of algorithm can determine what people are saying in public, just by lip-reading analysis.
On the other hand, this same algorithm can also enhance privacy by “shuffling” images and preventing other systems from doing facial recognition.
“Beer Geek. The Evil Ninja of Pop Culture. Life Coffee Scholar. Professional Internet Teacher. Meat Teacher.”