Deconvolution sharpening in Photoshop?

The Richardson-Lucy deconvolution deblurring and sharpening tool is widely used by astronomers (for example in correcting the blurred Hubble telescope pictures before the optics were fixed), but it does not seem to be used that much in general photography. I see that Raw Developer now offers it as an option and Roger Clark <http://www.clarkvision.com/imagedetail/image-restoration1/> has shown how it can be used to advantage in general photography, allowing him to enlarge twice as much as compared to unsharp masking techniques.

Are there any plans to incorporate this algorithm as a Photoshop plugin? I understand that the new smart sharpen used some kind of deconvolution kernel, but have no details on how it works.

FWIW, George DeWolfe, in his book "Digital Photography Fine Print Workshop" claims use of similar algorithms in his Optipix software. He also claims that CS2 Smart Sharpen filter with Lens Blur is similar as is the nik Sharpener Pro 2.0 Raw Presharpening filter. I have personally used none of these and am not familiar with them.

Larry

Maybe is it possible to do it with the custom filter? Or Reindeer Graphics's free plug-in: < http://www.reindeergraphics.com/index.php?option=com_content &task=view&id=34&Itemid=57> ?

Maybe is it possible to do it with the custom filter? Or Reindeer Graphics's free plug-in: < http://www.reindeergraphics.com/index.php?option=com_content &task=view&id=34&Itemid=57>
?

That's a good thought, but I wouldn't know what values to put in the matrix. Reindeer does have the Richardson Lucy algorithm in their Fovea Pro package, but it costs more than Photoshop and would be overkill for most of us.

I note that the custom filter is a convolution filter, whereas RL is deconvolution. Perhaps one would put in the inverse of the convolution matrix?

Wait for Mathias Vejerslev to show up in this thread, he has experienced several algorithms...

I've had a few images where I felt Focus Magic coupled with Photokit (Output Sharpen) did a better job of sharpening than did Photokit Sharpener alone (i.e. using both PK Capture and Output Sharpening). It may be worth your time to experiment with the free download which you can later purchase if you want.

<http://www.focusmagic.com/>

Ken

On May 9, 5:51 am, wrote:

That's a good thought, but I wouldn't know what values to put in the matrix. Reindeer does have the Richardson Lucy algorithm in their Fovea Pro package, but it costs more than Photoshop and would be overkill for most of us.

I note that the custom filter is a convolution filter, whereas RL is deconvolution. Perhaps one would put in the inverse of the convolution matrix?

Hi All,
A couple of notes about deconvolution. There is a distinction between deconvolution and blind deconvolution. For deconvolution, the blur function is in-hand, either by measuring it (e.g. taking an image of a single star) or modeling it using camera design (rarely ever done). When you use an filter, you can consider it as a very crude, but sometimes effective blind deconvolution.

The RL method is a deconvolution method that calls for having a blur function, also called a point spread function (PSF). Some people have made modifications to the approach such that it operates on a wider range of images, creating a Blind RL method. However, it is not the only approach. Maximum entropy methods are also effective and prevalent. Now, I am biased on this last point, but for convenience, speed, and effectiveness, its tough to top the results provided by SeDDaRA. Descriptions of blind deconvolution methods can be found at http://quarktet.com/BlindDecon.html, while many examples start at http://quarktet.com/Gallery1.html.

One caveat though, with any of these methods, the PSF must be preserved in the image for the techniques to work. If this information is lost due to noise or digital truncation, the restoration will not be very effective.

Jim at Quarktet

Bump...

I'm sure that the engineers considered most recent papers about sharpening when designing Smart Sharpen. It would be maybe wise to make a Features Request to have that algorithm added to that filter, if it is not already the case.

Smart Sharpens 'Remove lens blur' filter is a deconvolution filter, and so is its 'Remove Motion Blur' filter.

Smart Sharpens 'Remove lens blur' filter is a deconvolution filter, and so is its 'Remove Motion Blur' filter. Improvements could probably be done to both flters, if enough voices chime in (I'm all for highest quality, slow filters).

Yes it does seem to be some type of deconvolution filter, but how it compares to some other deconvolution methods such as the maximum entropy, Van Cittert, one-step gradient, and Richardson-Lucy iterative algorithms is not clear.

I presume that the Gaussian and lens blur are point spread functions associated with the deconvolution and that accuracy check box has something to do with the number of iterations. How best to use the smart sharpen filer is not well documented and it is not clear to me if I should try to master it or look for something better from the outset. Bruce Fraser wrote a whole book on sharpening and a couple of pages in the manual is not really enough to get a handle on the new method. In the final analysis, Bruce was not that impressed with the smart sharpen filter.

Anyway, the smart sharpen filter seems quite a bit different than the Richardson-Lucy algorithm that Roger Clark used in his example, where he tried various iterations, point spread sizes, and noise thresholds. His final filtration used 20 iterations and took 1.5 hours to perform on a 1.8 GHz Pentium 4 machine.

For the time being I think that I will put further consideration of these methods on hold until some one writes a good review of what is available and the settings that should be used.