Google brings machine learning to deliver sharper low res images

Slipping under the radar late last year was the potentially groundbreaking news that Google has come-up with a very clever way to deliver higher-quality versions of low resolution images. The technology is called RAISR (an abbreviation of Rapid and Accurate Image Super-Resolution since you ask) and it uses machine learning to produce what Google believes to be comparable or better results than other current super-resolution methods, while being between 10-100 faster at rendering as well. Whereas upsampling demands a larger file size with more pixels to attempt to get a higher quality image from a low-res file, it can still be quite ineffective at improving detail. RAISR works differently, using machine learning to train on pairs of images - one low quality, one high. It then finds filters that when selectively applied to each pixel of the low-res image, manages to deliver results that Google reckons is comparable in quality to the original higher-res image.