Oh. I've tried finding that exact publication - they took a video of a city scene with dense fog/clouds in the background. If you'd play the video you couldn't

Oh. I've tried finding that exact publication - they took a video of a city scene with dense fog/clouds in the background. If you'd play the video you couldn't see any shapes through the fog. At all. But after passing the whole video through reconstruction/de-noising, they've been able to recover the background (you could see s mountain, instead of dense fog). It is not surprising and it's a pretty standard result actually. Can be called by different names (sparse sensing, denoising, superresolution, ...) and done by different methods (BP, LASSO, ...), but it always comes down to having a model (with some priors, e.g sparsity prior!) and using that model to re-project the data from your measurement projection into the projection of some interest to you. I have no idea at what level you are, so I'd suggest this very well presented lecture (1 hour total): Compressed Sensing by Terence Tao http://www.youtube.com/playlist?list=PLC94A02A1218B24DF http://www.youtube.com/playlist?list=PLC94A02A1218B24DF if you want to learn more. Examples of image recovery are in the part 5, part 6.