Google unveiled prototype software called Rapid and Accurate Image Super-Resolution (RAISR) that pulls up a low-quality image and requests that the computer enhance the resolution. The machine learning software zooms into a photo without losing the picture’s quality and makes the appropriate adjustments.
RAISR works similarly to current methods of upsampling, the process that turns a small image into a large one by inserting new pixels into it. While traditional upsampling methods make the images bigger by filling in new pixel values using fixed rules, RAISR uses its methods according to the type of image it’s looking at.
The software makes use of what’s referred to as “edge features” – parts of a picture where the brightness or color gradient changes quickly and often indicates the edge of an object. The software analyzes its edge features and then selects and applies the most relevant filter from its collection. Adaptive unsampling is the result of zoomed images that are less blurry.
In the composite image below, the top portion is the original, low-resolution photo, and the bottom is the RAISR-enhanced version.
If we compare that to the image below, which shows the low-resolution picture on the left and the upsampled version on the right, we see the resulting photo is less pixelated, and its edges seem blurred and out of focus.
While RAISR is not available for download just yet, Google notes that its system requirements are modest enough for real-time photo enhancing on mobile devices.
Source: The Verge
Advertisement
Learn more about Electronic Products Magazine
