I found recently an article in which they describe a technique to extract high resolution images from a standard quality videos.
The technique exploits the fact that in a typical video the camera isn't still, and some parts of the scene get captured in more than one frame, allowing the software (as our brain does) to do some kind of "incremental enhancing" of the image by analyzing the frames that appear to be overlapping. If I understand well this is called re-sampling.
The output could be a "variable resolution still image" with some very high detail areas (where the camera passed more than once), and low resolution zones where there very few frames covering it.
This is the article: http://www.eecs.harvard.edu/~kalyans/research/snapshots/VideoSnapshots_TVCG12.pdf
Does anybody know about an open source project based on this idea? Thanks in advance.