If you look at old interview (first posts on this forum history) you'll see me mentioning deconvolution algorithms. This is them. Of course, he have exactly recorded camera motion path and few other variables set :-)
I've definitely seen that Patch match before....but that "Debluring" thingy is too cool. I wonder if this will port over to video one day. Would be cool to be able to fix those missed out of focus shots.
Does the camera need to provide orientation info for Adobe's unblur filter to work? That is what seems to be suggested. In that case, will cameras need to support this in hardware before photos can be unblurred this way?
@eoshd in the presentation he unblurs a random picture he found on google image search. i assume that additional camera data (trajectory/orientation) would not be available from that.
i don't think those files he is loading contains any camera movement data. yes he says he "loads some pre-selected parameters". this would seem to me to be the best plugin parameters he previously found to work with each image. you can see the plugin has three tabs on it probably with various settings that need to be tweaked to maximise algorithm performance depending on the type of image, and it would save time for the presentation to have those saved in a parameters file rather than going through and setting them all one by one.
after loading the settings file, he still has to press "analyse" for it to analyse the image to figure out the motion path (and the processing time is also dependent on how much of the image is selected for analysis), once analysis is finished then the motion path is shown. if the motion trajectory was already captured and being fed to the plugin via the settings file, you would expect that the motion path would be already available as soon as the settings file was loaded. unless it is an elaborate hoax to make it look like it is doing something.
also this program called unshake already does this http://www.zen147963.zen.co.uk/ with an explanation of how it works here http://www.zen147963.zen.co.uk/how/index.html it seems that unshake does some analysis to "guess" the best deconvolution, whereas i believe adobe has a more sophisticated way determine camera path from the image data.