MANILA, Philippines – You've probably seen it a few times in works of fiction: an investigator using high-tech software to clear up a blurred image and see who the suspect or criminal is. For many years, the method had been nothing but fictional fantasy.
But that soon may change with the progress that Google Brain, a research arm at Google, has made.
The division published a research paper that demonstrated a kind of artificial intelligence (AI) that was able to produce a clear photo of a face derived from a heavily blurred image consisting of 8 pixels by 8 pixels. A demonstration of what the technology was able to do is seen in the photo above.
The research paper, published on Thursday, February 2, revealed that the method made use of two brain-like information processing systems called neural networks: the "conditioning" network and the "prior" network.
In the first, the AI made use of a database of high-resolution shots, to which the low-resolution image can be compared and eventually matched. The AI downgraded the high-resolution shots into 8 pixel by 8 pixel images – and then identified which one matched the sample low-resolution image the closest.
From this, the system derived the basic form of the image. The "conditioning" network then steps in, fills in the details and sharpens the source image, adding new pixels along the way.
Ars Technica offered a simplified explanation of the process: "For example, if there's a brown pixel towards the top of the image, the prior network might identify that as an eyebrow: so, when the image is scaled up, it might fill in the gaps with an eyebrow-shaped collection of brown pixels."
At this stage, the technology can be relied on to give one an idea of what's there in a pixelated photo. While Google is making inroads in this crime fiction-inspired technology, it's not yet consistent nor mature enough to be used in pinning a crime on a suspect, based on its approximation. – Rappler.com