Google intelligent image scaling could mean 4K quality at HD data rates

Thursday, 19 January 2017

PC magazine has written about a new Google technology that shows very good results in scaling up lower resolution stills to higher resolution:

The new technique is called RAISR, which stands for "Rapid and Accurate Image Super-Resolution." [It] works by taking a low-resolution image and upsampling it, which basically means enhancing the detail using filtering. Anyone who's ever tried to do this manually knows that the end result looks a little blurred. RAISR avoids that thanks to machine learning.

Check out the article for examples of what RAISR can do.

Imagine this idea applied to ‘UHD’ video distribution. Store frames as 1920x1080 but with more colour information. To display 3840x2160 video, the algorithm could interpret the missing pixels and show an image with detail that even those very close to the display would find indistinguishable from real 4K.

Although this would be hard to do in real-time (in less than 1/60th or 1/120th of a second depending on frame rate), the the algorithm would need to interpolate less detail for moving video. There are limits to what the human visual system can discern. If video was slowed down or paused, the algorithm would have more time to produce higher quality 4K.

In practice 4K for broadcast is mainly for bragging rights, less than 1% of people sitting at a comfortable viewing distance are able to see the improvement over HD. It makes more sense to dedicate precious bandwidth to better quality pixels. This algorithm will then be able to scale the image up well if need be.