Wow ! These examples are amazing. I had no idea that deconvolution could work so well !
Congratulations Terry - this is a significant achievement ! I can't wait to see what happens when this PSF deconvolution is applied to the whole image. If I understand correctly, the pixel-perfect alignment *might* not be as critical as we first thought. FWIW I did a few experiments for "CPfind vs motion blur", exploring the idea that motion blur itself will create problems for CPfind accuracy. I created a simulated motion blur version of a sharp image in Photoshop ( 10 pixels at 45 degrees ), and then ran both through CPfind. If CPfind interpreted the motion blur correctly/consistently, the CP error distance should be (predominantly) 0.00 pixels. This did not happen. The CP error ranged by a factor of ten, eg from 0.93 to 9.32 pixels. I tested two more simulated motion blurs -- 10 px at 90 degrees ( vertical ) and 10 px at 0 degrees ( horizontal ). The CP error range in each case was also ten times, though the accuracy was best for horizontal blur ( 0.42 to 4.20 ), and worst for 45 degrees. I'm not really sure what to make of this, but thought I'd share anyway ! :-J On Apr 5, 2:10 am, "Terry Duell" <[email protected]> wrote: > On Mon, 04 Apr 2011 18:28:05 +1000, kfj <[email protected]> wrote: > > Well done! Once the simple test case works, you can at least hope more > > complex signals will yield as well - maybe not perfectly, but to a > > degree. Best of luck. I'll be off on a holiday - who knows maybe > > you'll have it all sorted by the time I'm back! > > A bit more progress. > I have used JohnG's example blurred and sharp images to extract an > overlapping pair (256x256). The alignment of the pair was a bit of guess, > done by eye in gimp by simply selecting a point in one image as the corner > of a crop area, then finding the same point in the other image. > A point spread function (PSF) was derived from this pair and then the PSF > applied to the blurred image to derive a new sharp image. > The result looks very good. > The attached files tell the story. > 'test-blur.pgm' is the original blurred greyscale sub-image > 'test-sharp.pgm' is the original sharp greyscale sub-image > 'test-newsharp.pgm' is the derived sharp sub-image > 'oct-do-deblurr' is the Octave script to do the sums > > The next step is to see if the same PSF can sharpen other parts of the > original blurred image. > > Cheers, > -- > Regards, > Terry Duell > > oct-do-deblurr > 2KViewDownload > > test-blur.pgm > 318KViewDownload > > test-newsharp.pgm > 427KViewDownload > > test-sharp.pgm > 284KViewDownload -- You received this message because you are subscribed to the Google Groups "Hugin and other free panoramic software" group. A list of frequently asked questions is available at: http://wiki.panotools.org/Hugin_FAQ To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/hugin-ptx
