http://www.bautforum.com/showthread.php/36641-Super-Resolution
There is a technical review, if you're looking for the mathematical details, available here. A lot of the details of implementation are still sort of the laboratory art of astronomers and not widely written up outside of specific applications. For example, it is a sort of superresolution to compare a well-sampled image of a suspected binary star to a known single star, and ask whether the pixel values are better fit by one or two (slightly offset) components. You can also google "drizzle processing" for a more general approach, improving image sampling by combination of multiple offset images of the same field (as done in the assorted Hubble Deep Fields and now a part of the Hubble Advanced Camera pipeline processing).
The diffraction limit is not a sudden cutoff - there is still some information in an image beyond the diffraction "limit", but progressively more attenuated on finer and finer scales. In superresolution, one is trying to use what's left of this information without being overwhelmed by the noise, which becomes more and more dominant toward higher spatial frequencies. This is less overwhelming in radio interferometry because the noise does not occur directly in the image, which is why radio astronomers have been happily doing deconvolution for forty years now and it remains a ticklish process for optical images. |
显身卡
