People may finally be able to enhance images that have very low resolutions. The folks at Google Brain is apparently making this happen via an impressive smart approximation software. Apparently, Google Brain was able to turn grids of pixels into a more comprehensible approximation of human faces. Here's how it works.
How The Google Brain Team Is Able To Enhance Blurry Images
As per Gizmodo, the best method of increasing details on low-res photos is through neural networks. Even if you are stuck with just the pixels, deep learning can help you add in details in a manner that is commonly dubbed as "hallucination". Technically, it means that to get a clearer image of a pixelated photo, a "guessing" software must be used. This software will make use of all the info it has learned from a variety of images to give its best approximation of what was actually captured in the pixelated image.
Google Brain Team's "Image Enhancement" Advancement
Google Brain has recently shared the results of their project and despite the fact that the resulting images look monstrous, they are actually very impressive. Take note that the original data used by the team were mere 8x8 pixelated grids, so having an understandable resulting image is pretty amazing. It's worth pointing out, however, that the end results are simply products of the software's imagination. Meaning, the computer just did its best to provide the best results, so it's important to note that the outputs may not be able to identify an exact person as the software isn't smart enough to magically come up with an identity. Needless to say, this tool could still help detectives with their jobs just as long as other relevant equipment or data is being used along with it.
Unfortunately, though, as per CNET, Google considers this project as a research exploration and it has no plans of using it commercially anytime soon. It is still, however, an interesting creation that has only been made possible in sci-fi movies. Hopefully one day, Google and other tech companies will be able to commercialize this feature to help different industries with their day to day needs.
