Inter BEE 2010: NEC Exhibits High-Speed Super-Resolution Technology Based on Multiple Images.
NEC (booth #5414) featured a reference exhibit of super-resolution technology which generates HDTV images from multiple images within an SDTV resolution stream.
Super-resolution is a technique by which an image with higher resolution than the original is generated through mathematical processes. Conventionally "super-resolution" referred to when multiple input images were used to generate an image with higher resolution than what was input. But more recently, the same name is used when TV receivers apply resolution-improving processes in real time, in which case the input is on a single frame basis.
With their new technology, NEC has returned to basics by using visual information from multiple images (specifically 3 frames before and after the target frame for a total of 7 frames) for its super-resolution process. In order to accommodate rapidly moving images, motion vectors are used to sync the location of the subject. As a result, it became possible to apply super-resolution processing even to moving parts of the image, whereas in the past it application was only possible to stationary parts of the image.
According to NEC booth staff, What's more, algorithm running speed has been greatly increased, and according to NEC booth staff, the exhibited device is capable of processing "at 10 times real time". This is using only a CPU, so by incorporating a GPU even further acceleration can be anticipated.
NEC says this technology can be used to convert entire archives to HD resolution.
NEC's super-resolution technology for converting SD content to HD is applicable even to images with movement.