Google has officially announced that their newly developed RAISR algorithm, based on machine learning and made for saving bandwidth and preserving or enhancing images, will be making its debut in a public product with a rollout to Google+. The algorithm, whose name stands for “Rapid and Accurate Image Super Resolution”, was made with machine learning, with the goal of being able to resize images with little to no loss in fidelity. It will be having its first public run with this rollout to Google+, where the service will request an image that’s only one fourth of the original size, then upscale it on-device using RAISR, producing an image that’s comparable to the original. This will result in a data savings of about 75% for the user.
The way RAISR works sounds simple, but is indeed a modern marvel made possible by recent advancements in machine learning. When a small image or thumbnail is blown up, it ends up with gaps in the pixels. Traditionally, a computer will fill in those gaps with adjacent colors, resulting in artifacts and a blocky appearance, among other defects. RAISR was trained by giving it about 10,000 sets of images; one of each set was small, and the other was large, but with the gaps filled in lovingly by hand to be as faithful as possible to a higher-resolution original. Because of this training, RAISR is able to more accurately fill in the gaps in a supersized image.
Google first announced RAISR back in November, meaning that the protocol has only been around for a few weeks and is already rolling out on a primetime basis. Google did indicate, however, that it will see a limited rollout on Google+ at first, and if things go well, it will roll out further at a later date to be announced. In their blog post on the matter, Google said that they are planning on thinking up other applications for RAISR in their users’ day to day operations, which will likely involve similar implementations to what we’re seeing here. RAISR could, of course, have many different applications going forward, so we will have to wait and see what Google has in store, or if they open it up to developers.