Could AI make all our old blurry photos clear?

In this example, the algorithm reworked a 64x64 pixel image into a near perfect 1024x1024 pixels. — AFP Relaxnews

Google has revealed that it is working on a new technology that will considerably improve the rendering of certain low-definition images. Algorithms allow thumbnails to be transformed into perfectly sharp high-resolution images. Deployed on a large scale, this type of technology could enable countless poor-quality pictures to be ‘saved’.

An optimisation algorithm, called SR3 (for “Super-Resolution via Repeated Refinements”) is responsible for improving the image through various successive steps, gradually adding Gaussian noise or deleting certain details. Over the course of the experiments, thanks to machine learning, the image is then refined and Google’s teams are now able to reproduce perfectly clear high-resolution images from very low-resolution shots.

The results on faces are quite impressive, especially when switching to resolutions much higher than those of the basic low-resolution image. Google has uploaded a multitude of examples of 32×32 or 64x64 pixel photos perfectly enlarged to 1024×1024.

Who hasn’t scoffed sceptically while watching a TV show or movie where ultra-powerful machines allow investigators to easily zoom in, in a perfectly clear way from relatively mediocre images? Well, this experiment carried out by Google suggests that one day this may become a reality.

More concretely, this technology could also one day offer anyone the possibility to enlarge and sharpen up old digital photos or even scans of old family photos. – AFP Relaxnews