Wondering how hard it would be to convert the functions for IM.
Amazing demo: DeBlurring Images..
Amazing demo: DeBlurring Images..
http://yuzhikov.com/articles/BlurredIma ... ation1.htm
Wondering how hard it would be to convert the functions for IM.
Wondering how hard it would be to convert the functions for IM.
-
fmw42
- Posts: 25562
- Joined: 2007-07-02T17:14:51-07:00
- Authentication code: 1152
- Location: Sunnyvale, California, USA
Re: Amazing demo: DeBlurring Images..
see http://www.fmwconcepts.com/imagemagick/ ... urier.html and my FFT scripts at the link below.
Last edited by fmw42 on 2012-10-22T14:20:06-07:00, edited 2 times in total.
-
NicolasRobidoux
- Posts: 1944
- Joined: 2010-08-28T11:16:00-07:00
- Authentication code: 8675308
- Location: Montreal, Canada
Re: Amazing demo: DeBlurring Images..
I hear that Vladimir Yuzhikov has released his SmartDeblur program under the GPL at https://github.com/Y-Vladimir/SmartDeblur
-
NicolasRobidoux
- Posts: 1944
- Joined: 2010-08-28T11:16:00-07:00
- Authentication code: 8675308
- Location: Montreal, Canada
-
hjulenissen
- Posts: 18
- Joined: 2012-10-05T03:44:29-07:00
- Authentication code: 67789
Re: Amazing demo: DeBlurring Images..
Whenever I see such amazing examples, I am struck by sceptisism:
1. Is the image degraded by a real, relevant blurring (such as out of phocus lense, camera movement), or by a synthetic degradation that satisfy some ideal property (such as space-invariance, noise-free etc)
2. How much manual work is needed to estimate/guide an algorithm in finding the optimal PSF estimate?
3. Even though the ability to "zoom and enhance" until readability is amazingly improved, the subjective qualities of the result leaves a lot to be desired: it looks definately "processed".
In my own photos, I may want some degree of blur in out-of-focus parts of the scene (googled example http://net.onextrapixel.com/wp-content/ ... assdew.jpg), and for stuff that is moving fast (googled example: http://api.ning.com/files/UKR6RTdFcZk1r ... tos16.jpeg ). It is a part of my "artistic reportoire" if one may use fancy terms. I may also operate at the limit where image noise is barely tolerable. So the question is if fancy deconvolution based methods can be used, automatically or semi-automatically to make my images "better", both in the objective sense of "less error compared to what the image file would be like if I had used a 20kg stand and $10000 lab-grade lenses and had nailed focus" and in the subjective sense "looks more real/prettier to me".
Seems to me that one would ideally have a complete description of camera PSF at all apertures, focal lengths, distances, wavelengths etc _and_ some non-linear, adaptive thingy that compensates for the deviations from the general behaviour one might encounter in any given image, then recover assumed image information weighted against artifacts/noise amplification in some mean-squared-error (or more elaborate HVS sense). Then you need to be able to blend recovered image with the unprocessed image so as to keep artistic intention/satisfy individual taste for enhancement vs artifacts.
Perhaps easier (but less academically interesting) to buy the more expensive camera gear and improve ones skill in using it
-h
1. Is the image degraded by a real, relevant blurring (such as out of phocus lense, camera movement), or by a synthetic degradation that satisfy some ideal property (such as space-invariance, noise-free etc)
2. How much manual work is needed to estimate/guide an algorithm in finding the optimal PSF estimate?
3. Even though the ability to "zoom and enhance" until readability is amazingly improved, the subjective qualities of the result leaves a lot to be desired: it looks definately "processed".
In my own photos, I may want some degree of blur in out-of-focus parts of the scene (googled example http://net.onextrapixel.com/wp-content/ ... assdew.jpg), and for stuff that is moving fast (googled example: http://api.ning.com/files/UKR6RTdFcZk1r ... tos16.jpeg ). It is a part of my "artistic reportoire" if one may use fancy terms. I may also operate at the limit where image noise is barely tolerable. So the question is if fancy deconvolution based methods can be used, automatically or semi-automatically to make my images "better", both in the objective sense of "less error compared to what the image file would be like if I had used a 20kg stand and $10000 lab-grade lenses and had nailed focus" and in the subjective sense "looks more real/prettier to me".
Seems to me that one would ideally have a complete description of camera PSF at all apertures, focal lengths, distances, wavelengths etc _and_ some non-linear, adaptive thingy that compensates for the deviations from the general behaviour one might encounter in any given image, then recover assumed image information weighted against artifacts/noise amplification in some mean-squared-error (or more elaborate HVS sense). Then you need to be able to blend recovered image with the unprocessed image so as to keep artistic intention/satisfy individual taste for enhancement vs artifacts.
Perhaps easier (but less academically interesting) to buy the more expensive camera gear and improve ones skill in using it
-h