Google's new artificial intelligence project shows promise for the world of image enhancing. The search engine giant's neural networks can create an almost realistic image just from an eight by eight pixels grid image.
What was once a famous TV trope may finally become reality. Zooming in and enhancing images may become possible with the new Google AI project if the company does continue to improve upon its research and experimentation.
Google's Brain research team used two neural networks and trained them to recognize and enhance the images by feeding them photos of bedrooms and faces of celebrities, CNET has learned. They were then given a super low-resolution image to test their capabilities.
The first neural network figured out how the small set of pixels can be transformed into a better image. It did this by downsizing other images to the same grid and comparing them to find a match.
Google Brain's second neural network then worked to improve upon the first one's image by adding finer details. If a set of pixels looked similar to an eye or a lip, it would find similar looking details and try to paste them onto the image.
Human observers that were shown the celebrity images and Google Brain's enhanced images were fooled 10 percent of the time, Ars Technica has learned. The bedroom images test saw an even more impressive rate of 28 percent with 50 percent being the perfect score.
Google Brain's image enhancing tech could still be used as a guide but not an absolute one for police force and investigators. The search engine giant told CNET that there are currently no plans to incorporate the tech with their products.
Unfortunately, the technology cannot be used by law enforcement at its current state. It is not perfect and there are more improvements to be made. In addition, the enhanced images of human faces may lead to mistaken identity which can cause more problems in the long run.
Learn more what Google Brain can do: