A scaled gradient projection method for constrained image deblurring

[hartford]

I came across this:

http://www.iop.org/EJ/abstract/0266-5611/25/1/015002
A scaled gradient projection method for constrained image deblurring

The gist is that this article describes how to deblur an image.

The article costs US$30. I'd buy it but I doubt that I could
translate the math into a useful plugin.

Perhaps someone with better math and computer language skills might find this interesting.

[hartford]

I did find a source on the same topic: A Scaled Gradient Projection Method for Constrained Image Deblurring:
http://cdm.unimo.it/home/matematica/zanni.luca/B_Z_Z_IP2008.pdf

In fact, there seems to be a number of publications available from Italian sources in pdf format.

So, I'm wondering if this has been explored and if so what was the verdict.

[hartford]

It does come to mind that this was discussed a while back and that camera parameters were needed which would make the above useless.

More learnerd please jump in.

[hartford]

Ok. Topic dead.

[Mark Rejhon]

It does come to mind that this was discussed a while back and that camera parameters were needed which would make the above useless.Without looking at the paper, how much computational power is needed to deblur one image? How many camera parameters are needed?

In theory, would it be possible to run this deblur algorithm with many random camera parameters, and plug-in some traditional digital autofocus algorithms (edge contrast detection) to automatically fuzzy-logic your way to the original camera parameters? That may mean thousands of test frames before the sharpest frame is found. And to speed things up, only certain parts of the image (such as the center) could be processed on -- much like traditional digital autofocus algorithms.

Ungodly amounts of computing power -- but with teraflops in a GPU available this year -- maybe theoretically pratical, if camera parameters are a finite-enough set to fuzzy-logic towards original parameters? Once found, the same camera parameters can then be used for the next frame in the video stream, to save a lot of CPU power.

Anyway, maybe not pratical, but never say impossible. Some of today's algorithms are doing miracles.