Google takes a step further to achieve impossible CSI technology

Re-creation of a pixelated image to find the suspect is quite common in any crime series but is not as easy as it seems. We haven't had much success in that sphere. Google, as usual has come to our rescue with its Google Brain team using a  "zoom and enhance" technique.

Courtesy: Google Brain

As Ars Techinca reports, Google basically uses neural network to fill in details of a low resolution image(8 pixel by 8 pixels) to get an approximate view of what the original image may look like. It uses a "conditional" neural network to map the low resolution image to other high resolution image to get a basic idea of the details. It then uses a "prior" neural network to make the details credible.

Courtesy: Google Brain

To be clear, it is not a magical recreation ( the image is not real) since you cannot add more detail than already present but it can come fairly close to the real picture. A totally useless(8 pixel by 8 pixels) picture has some details now (maybe inaccurate) but it definitely prompts what might be there.

What to expect?
It is not the first time we see the use of AI and neural networks in image manipulation. That's how Google image compression works. It has still not reached perfection but is a big step towards accurate suspect identification. It may be used for cleaning up finer details when the low-resolution images are blown up to higher pixels. It does make the image more reasonable.

P.S: You can read the whole research paper here.