Increasing resolution of existing photos has traditionally been done by a combination of extrapolation of the existing images, followed by some form of more-or-less smart sharpening. The Max Planck Institute for Intelligent Systems, Tübingen, Germany, has developed a new machine-learning based system named Enhancenet-PAT, which shows much greater success in increasing photo resolution using automated texture synthesis. Here's a talk (14-minute video) that explains this in more depth.
