Google shows the power of machine learning by enhancing blurry, low-res pictures

We don’t need to further emphasize just how important AI and machine learning are for Google. Whether it be in its cloud services or inside of its Assistant-powered devices, like the Pixel phones and Home, we know that there is substantial room for improvement through technology, and the Mountain View firm is all for it.

Now, it seems, it’s low-res and blurry pictures’ turn…

The last magic trick out of the firm’s hat is a prototype software called RAISR, which stands for “Rapid and Accurate Image Super-Resolution”. Unlike traditional upsampling, which simply enhances an image by filling it with more pixels, RAISR utilizes machine learning algorithms to do the same thing in a more intelligent way.

For instance, one of its keystones lays in paying attention to so-called “edge features”: when color gradients and brightness change dramatically within a short span, that usually means that that’s the edge of an object. Bolstering an image with the traditional method may up its resolution, but the results, as shown in the sample below, usually end up being blurry and out of focus.

RAISR’s intelligent retouch, instead, manages to retain the original shapes more accurately, overall resulting in an image that simply results better looking. Says Google:

“In practice, at run-time RAISR selects and applies the most relevant filter from the list of learned filters to each pixel neighborhood in the low-resolution image. When these filters are applied to the lower quality image, they recreate details that are of comparable quality to the original high resolution, and offer a significant improvement to linear, bicubic, or Lanczos interpolation methods.”

This is the full process:

Here are two additional pictures, which show the difference between the original image (above and left, respectively) and the one gone through RAISR’s filtering (below and right).

There’s no indication that something like a dedicated app is on its way, but the Google post’s conclusion seems to be hinting that it may reach end users in some way, perhaps not too far from now. It reads:

“For example, in addition to improving digital “pinch to zoom” on your phone, one could capture, save, or transmit images at lower resolution and super-resolve on demand without any visible degradation in quality, all while utilizing less of mobile data and storage plans.”

We just hope it won’t take too long.