Terry, I modified the MATLAB script to only run their algorithm and use the provided motion blur blur kernel on a full size image. Here are the results when run on an 18MP photo I had:
Original: http://i.imgur.com/rRwaKbc.jpg Blurred: http://i.imgur.com/NyGg6JN.jpg Deconvoluted: http://i.imgur.com/fCeOTGy.jpg The script takes the original image, applies the blur kernel, adds Gaussian noise to both the blurred image and the blur kernel, and then runs the deconvolution. It took an hour and a half, maxing out all four cores of my 4.2GHz Ivy Bridge processor. If you have any images you want me to test, I'd be happy to do so. We just need a good way to estimate an image's blur kernel. -Matthew On Thursday, October 3, 2013 7:16:32 PM UTC-4, Tduell wrote: > > Hello Matthew, > > On Wed, 02 Oct 2013 09:41:06 +1000, Matthew Petroff > <[email protected]<javascript:>> > > wrote: > > > Terry, > > > > I read the paper, and I don't think you'll be able to use your existing > > images. The technique is based on first calibrating a lens to find > > spatially varying point spread functions via a patch-based method and > > then > > using the known blur kernel to correct subsequent images. Without > > properly > > calibrated point spread functions, it won't work, and none of the > > downloads > > provide their calibration target or code to generate the point spread > > functions. It also only corrects for blurs inherent in the lens optics, > > > not motion blurs. > > > > I have a bit more of a think about this, and I am inclined to think that > the method may be applicable to the correction of any image, if one can > estimate the PSF. > If I understand it correctly, their approach is to derive, by test, a PSF > > for the lens, then apply that PSF to correcting the image . > If one has other methods of deriving the PSF from the subject image (or > from a sharp and blurred pair as might be the case for pano work) then > one > could use their method to apply that PSF to correcting the image. > A look in the "kernels" directory of the code package shows a kernel > "motion_blur.png" and some others that have the characteristics of motion > > blur. > The above is a tad academic right now as the only means I have of testing > > this is to use the code package with Octave. > As you have said, some of the Matlab functions that are used are not > available for Octave, and I'm not sure how I will go getting over that > hurdle. > > Cheers, > -- > Regards, > Terry Duell > -- A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ --- You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/hugin-ptx/acd7e249-a6dd-4afc-8e0c-6152a7c7e8d0%40googlegroups.com. For more options, visit https://groups.google.com/groups/opt_out.
