Recently, the team behind Google Brain made some very serious innovations in what concerns increasing the resolution of images. The technology we all saw in detective movies where they look at very low quality photos of suspects and they ask the computer to “zoom and enhance” is now real. All right, almost real. Google managed to explore that technology and turn small pixelated images into clearer ones.
The science behind this technology
Still, how does this technology work? It all has to do with neural networks. Thanks to those, the level of quality in a low-resolution image can be enhanced. A photo contains only pixel information but thanks to deep learning, we can now add detail to those images which are called “hallucinations”. A computer will be able to do this. Basically, it will guess details of the image based on what it has previously learned from studying other images.
The Google Brain team recently made public its researches. It shows how small and pixelated images can gain details through some steps. For now, the final results look rather horrifying, like very deformed people. However, the idea is not for them to look like catalog images. The idea is to transform them into more detailed photos. Which this new technology did successfully.
No plans to actually use the technology
How does Google Brain do all of this? Firstly, it downsizes higher resolution images to 8×8 and tries to match them. Then, the “prior” network uses something called PixelCNN. This allows the addition of realistic details to the small initial image. After all those steps, the initial source image is up-scaled. Only then, the system adds the new pixels in such a way that it matches the image. Finally, for the creation of the final image, the results from the two neural networks are added together. What comes out is the final, more detailed image. It sounds rather difficult but the team at Google Brain says that it is not that hard actually.
The bad news? They are saying that this technology will unfortunately not be used by the police the same way we see in movies. At least for the moment. According to them, this was only a research exploration. Still, who knows? Maybe in the future, this will serve as the basis for another technology which will reveal the people behind pixelated images.
However, some are saying that this technology should not be used. For once, authorities might want to still use it and because the results are not precise enough, they could end up arresting innocent people, the way it happened in this case. Apart from that, some artificial intelligence based software is known to be, as weird as it may sound, racist. AI often does not recognize different races of people. So, this might also be a big problem. All in all, it is nice that the people at Google are coming with such innovations which may prove very useful in the future. Until then, we will stick to the old ways of catching criminals.
Image source: here