PKS Sharpening Shoot-Out

[Mark D Segal]

Try right-clicking, Save Target As... or your browser's equivalent. A 16-bit TIFF isn't going to display properly in IE or any other browser I know of. Save the file and then open it in PS.
[a href=\"index.php?act=findpost&pid=145147\"][{POST_SNAPBACK}][/a]

Thanks Jonathan. That worked - should have thought of it myself!

[bjanes]

It's not true that deconvolution cannot produce completely out of focus detail.
Have a look at http://www.bialith.com/Research/BARclockblur.htm
It seems from comments from Dr Tadrous and others that commercial photo programs are useing mainly the Van Cittert algorithm which is not much better than USM. Bill, I'm playing with the Lucy -Richardson algorithms which are available. I've yet to get onto Landweber, which has impressive results in Tadrous's work. . Perhaps someone could help. I'm geting email bounces from astrophysicists.
Adobe's Smart Sharpen is onto something, but it's a bit crude at the moment. It does combine USM with some sort of deconvolution. You need to excange the Gaussian Blur option for Lens Blur and check Advanced. I think the two approaches to sharpening should be available seperately as they do completely different things. To really make use of deconvolution you need a lot of iterations, which would be disasterous with USM. USM is a perceptual edit , and does not reconstruct wavefronts.
Cheers
Brian
www.pharoseditions.com.au
[{POST_SNAPBACK}][/a]

Brian,

Thanks for the link. The degree of defocus in those examples is greater than is normally encountered in reasonable careful general photography, but the results are impressive. I note that a PSF was derived from a test shot, but in most routine photography, no such shot is available.

[a href=\"http://www.clarkvision.com/imagedetail/image-restoration1/index.html]Roger Clark[/url] is an astrophysicist and photographer who demonstrated some good results with an Adaptive Richardson-Lucy Iteration on his web site. He used an astronomical program, ImagesPlus. It costs US $200 and might be worth the price to a general photographer just for the sophisticated image processing options.

Bill

[sergio]

I just downloaded the demo and just for capture sharpening that is the only thing I've done til now, the results are very good. I used the recommended settings of blur 2. After I just ran my regular PKS edge1, backed it off to 80% and results are great.

[marcmccalmont]

Marc, what are the differences of treatment between the three shots?
[a href=\"index.php?act=findpost&pid=145219\"][{POST_SNAPBACK}][/a]

first one is no sharpening (the original file) the second 1 pass the third 2 passes
Marc

[ejmartin]

The last time I ran a comparison of different capture sharpening methodologies (I have Focus Magic, PK Sharpener, and CS3), my sense was that FM generated unwanted artifacts to a greater degree than the other two, while Smart Sharpen (lens blur, advanced mode with appropriate rolloff of highlight and shadows) beat PK Sharpen for more precise fine detail and less noise.  The one method I've not yet explored fully among these is the use of the sharpening tab in ACR 4.2, however I shy away from this possibility since the sharpening is applied globally and there is insufficient control of masking (though I haven't explored the possiblities of using different ACR sharpenings as smart layers and masking that in CS3).

IMO the place where PK Sharpener really shines is the output sharpening for print (Smart Sharpen wins once again for sharpening for web presentation).

[DarkPenguin]

I just use ACR 4.2's sharpening.  Works great.  Taxes few brain cells.

Prior to that I used the canon recommended 0.3 0 200%.  That taxed virtually no brain cells but was okay at best.  (Although it seemed to work pretty good on the sample image here.)

I've always thought the real work in sharpening was in the output sharpening.  That be voodoo.  (Perhaps if I was willing to blow $99 on PK it wouldn't be but I have to draw the line somewhere.)

[Brian Gilkes]

Brian,

Thanks for the link. The degree of defocus in those examples is greater than is normally encountered in reasonable careful general photography, but the results are impressive.



No problem. I have seen Roger Clark's foxy work.
ImagesPlus is very sophisticated. Infinitely more so than Focus Magic or Focus Fixer. Interestingly they give Van Cittert a good rap. There needs to be a lot of work done to dovetail this stuff into photographers workflow. If Roger's analysis is correct a 22 MP 1Ds III would turn into 88MP just by restoring lens information.
Jonathan's workflow seems correct. Use iterative convolution algorithms on capture and USM before exporting sized file. This is basically what I do , but using RAW Developer for capture sharpening and large radius USM or edge filters in the creative middle.   This works OK on Mac. Most of the astro stuff is Win,including ImagesPlus and I'm not sure if it wouldn't slow too much on simulation. Parallels could be OK.
Cheers,
Brian
www.pharoseditions.com.au

[Schewe]

If PK wants to remain competitive, I think that they also might have to move beyond a USM based approach. I have asked Jeff if they are working with these methods, but thus far I have no response.
[a href=\"index.php?act=findpost&pid=145196\"][{POST_SNAPBACK}][/a]

Bruce and I both worked with Jeff Chien from Adobe who was primary engineer (along with a few others) when they developed Smart Sharpen. Bruce's problem was that while deconvolution algorithms may have a lot of promise when the optical defects can be known and an algorithm specifically designed to remove a known defect, it's not clear that "general" deconvolution algorithms can be derived without knowing the exact aberration it will be used against.

That was Jeff's problem with Smart Sharpen...he actually showed us a 50 pass iteration that could effectively remove an applied gaussian blur, but he first needed to know the EXACT PSF that could attack the specific blur. Once known, multiple passes could essentially rebuild a soft image into a sharp image with little or only small defects...

Using deconvolution algorithms for removing specific camera shake is another area of interest because you can actually measure the shake involved and the line of bias and then go about removing that camera shake from the image. But, again, this is neither easy nor quick.

Smart Sharpen has both a Lens Blur as well as a Motion Blur mode...but there are no parameters to adjust the Lens Blur mode and you can't combine modes without separate runs...

Bruce had determined, after Smart Sharpen shipped, that there was no compelling reason to adapt PKS to incorporate it...he actually was more interested in DxO and used that earlier last year and was pleased because it offered a method of fixing a lot of lens defects all in one fell swoop. But I'm not sure if he thought of fixing lens defects in the same light as doing capture sharpening. He would run images through DxO and output a linear DNG and then run through Camera Raw for tone & color and finally used PKS for capture sharpening through to output sharpening.

But his time with DxO was rather short and came after finishing his first edition of RW Image Sharpening.

I also note that one of the reasons (excuses) that Hassleblad gave for taking DNG out of the camera output was that there wasn't an easy way to transfer the lens data to enable lens defect correction. Nikon also makes that claim for NEF and Nikon Capture. Canon, at PPE next week is due to release a new version of DPP which will now also supposedly read lens data and attempt to correct for some lens defects. I don't know to what extent Canon is going down that hole however.

All of this research is all very interesting, but to put something in a shipping product that can have a useful impact given the wide range of shooting conditions and lenses that photographers use is another thing. And correcting for very specific lens defects is very complicated as the potential matrix of defects is large–particularly zoom lenses whose defects vary along the focal length make something like DxO seem like a more logical solution. But, again, that still doesn't mean it's a "capture sharpener" per se...

[Gregory]

I wonder. with all the processes some photographers apply to their photos to get the perfect product, do they have any time to take photographs?

I bought CS3 so that I could use PKS because it simplifies the sharpening process for me. I don't need to think about or understand the process so much, and the printed/web photos are more than good enough for my needs. (now if only I could script the resizing/cropping steps...)

on the other hand, if a shipping product could reverse the camera shake in my 1DM3 images and could be incorporated into my (Aperture) workflow, I'd take a serious look at it.

regards,
Gregory

[luong]

He would run images through DxO and output a linear DNG and then run through Camera Raw for tone & color and finally used PKS for capture sharpening through to output sharpening.

That's  exactly my standard workflow, except that I save the master file without sharpening. I then use PKS most of the time, with mostly two exceptions: if I need stronger sharpening, or if I need to upres considerably.

[marcmccalmont]

It  is interesting that focusfixer reads the exif data to determine camera type and aperature ?
Marc

[Mark D Segal]

I just use ACR 4.2's sharpening.  Works great.  Taxes few brain cells.

Prior to that I used the canon recommended 0.3 0 200%.  That taxed virtually no brain cells but was okay at best.  (Although it seemed to work pretty good on the sample image here.)

I've always thought the real work in sharpening was in the output sharpening.  That be voodoo.  (Perhaps if I was willing to blow $99 on PK it wouldn't be but I have to draw the line somewhere.)
[a href=\"index.php?act=findpost&pid=145241\"][{POST_SNAPBACK}][/a]

Dark-Penguin,

My issue with using sharpening in the raw converter is one of control after the fact. If I sharpen there, then render the image, then say I do some other stuff including local contrast enhancement and find I don't like the resulting package, I may need to unscramble the omelet by reverting to the raw image and re-doing the sharpening, re-rendering and re-doing whatever else I had already done in Photoshop. Frankly, I'm not sure why it's worthwhile having a sharpener in a raw converter altogether, especially as PKS offers a layer-based solution at all stages. I can see it for Lightroom which aims to be a self-contained workflow for all the real basics, so it is needed there, but Camera Raw is a (increasingly powerful) prelude to Photoshop.

Turning to your "voodoo" on Output Sharpening, if you find this voodoo, then you are a good candidate for PKS. Believe me I have no commercial relationship with PixelGenius whatsover - you're just hearing this from an intensive user. It has that combination of quality and ease/flexibility of use that just makes it a compelling piece of software.

Because sharpening is such an absolutely critical part of the whole image creation process, it is the LAST place I personally would draw the line for spending on plug-ins. (There are many others far more expensive and less useful.) Whether one uses this or some other solution for output sharpening, the key thing is to get it right. One of the first things that always hits me when I examine photographs in galleries or craft shows is whether they have been under or over-sharpened. It is the first tell-tale of craftmanship in print processing, because it so obviously affects the overall photographic quality of the image.  One will see this before noticing, for example, whether the hue of a colour is several degrees to the right or left of where it should be.

[ejmartin]

Bruce and I both worked with Jeff Chien from Adobe who was primary engineer (along with a few others) when they developed Smart Sharpen. Bruce's problem was that while deconvolution algorithms may have a lot of promise when the optical defects can be known and an algorithm specifically designed to remove a known defect, it's not clear that "general" deconvolution algorithms can be derived without knowing the exact aberration it will be used against.

That was Jeff's problem with Smart Sharpen...he actually showed us a 50 pass iteration that could effectively remove an applied gaussian blur, but he first needed to know the EXACT PSF that could attack the specific blur. Once known, multiple passes could essentially rebuild a soft image into a sharp image with little or only small defects...

Using deconvolution algorithms for removing specific camera shake is another area of interest because you can actually measure the shake involved and the line of bias and then go about removing that camera shake from the image. But, again, this is neither easy nor quick.

Smart Sharpen has both a Lens Blur as well as a Motion Blur mode...but there are no parameters to adjust the Lens Blur mode and you can't combine modes without separate runs...

[a href=\"index.php?act=findpost&pid=145259\"][{POST_SNAPBACK}][/a]

This is interesting.  How do you translate a PSF into an algorithm for a sequence of parameter inputs to iterative applications of smart sharpen?

[ejmartin]

I find the application of a deconvolution technique to remove the effects of the anti-aliasing filter to be most interesting. The trouble with deconvolution techniques is in obtaining a point spread function (PSF) that describes the mechanism that produced the blur so that the deconvolution can undo it. I presume that the FM assumes a Gaussian function with a variable radius. The PSF required for removing the effects of an anti-aliasing filter might be different.

There are interactive deconvolution filters that allow the PSF and deconvolution to be observed in real time. An example is Fovea Pro from Reindeer Graphics.

Focus Fixer is another deconvolution method. They promise a White Paper that will describe the limitations of USM, "Has USM had it's day". They think that it has. If PK wants to remain competitive, I think that they also might have to move beyond a USM based approach. I have asked Jeff if they are working with these methods, but thus far I have no response.

Bill
[a href=\"index.php?act=findpost&pid=145196\"][{POST_SNAPBACK}][/a]

I agree that there is a lot of room for improvement in sharpening, and deconvolution is an intriguing possibility.  It works well in astrophotography, where in every image you have a lot of point sources available to determine the point spread function.

Wouldn't any deconvolution method be best applied before demosaicing the raw image?  It seems to me that otherwise, any algorithm has two distortions of the signal to contend with -- the blur introduced by the lens, and the filtering of that signal through whatever interpolation algorithm is being applied.  Furthermore, many interpolation algorithms are quite nonlinear in their attempts to preserve edges, reduce moire, etc.

[lllusion]

That's  exactly my standard workflow, except that I save the master file without sharpening. I then use PKS most of the time, with mostly two exceptions: if I need stronger sharpening, or if I need to upres considerably.
[{POST_SNAPBACK}][/a]

That was my workflow, but I found the DxO interface and output functions user unfriendly. Instead, I've created objective settings in LR for removing all vignetting (I used an opal diffuser and a specific methodlogy to come up with settings that limit subjectivity regardless of the lighting in the image.), and then run [a href=\"http://epaperpress.com/ptlens/]PTLens[/url] in an action to remove distortion. This workflow also means that the demosaicing gets done in LR instead of DxO, which I found to alter colors and WB in a manner I wasn't fond of. It also simplified my workflow.

[Schewe]

Wouldn't any deconvolution method be best applied before demosaicing the raw image?  [a href=\"index.php?act=findpost&pid=145325\"][{POST_SNAPBACK}][/a]

That's DxO's argument because that's what they do...and it's hard to argue against...once a capture has gone through some sort of demosaicing, the Bayer array is gone and the interpolated result would add even more complexity to the deconvolution algorithm requirements...

[Jonathan Wienke]

I wonder. with all the processes some photographers apply to their photos to get the perfect product, do they have any time to take photographs?

I've shot about 130,000 frames over the last 5 years. You?

[Marco Ugolini]

PKS appears to be USM-based
[a href=\"index.php?act=findpost&pid=145116\"][{POST_SNAPBACK}][/a]

Isn't PKS, in essence, a series of finely-tuned actions that make use of the tools that already exist in Photoshop?

In that case, I would not expect it to use any technologies outside the range offered by that application.

Marco

[Mark D Segal]

Isn't PKS, in essence, a series of finely-tuned actions that make use of the tools that already exist in Photoshop?

In that case, I would not expect it to use any technologies outside the range offered by that application.

Marco
[a href=\"index.php?act=findpost&pid=145358\"][{POST_SNAPBACK}][/a]

Marco, that's correct, but referring back to his remark and as Jonathan has I'm sure by now observed, it uses more of those tools than USM alone, e.g. layer masks, blend modes and other filters.

[Mark D Segal]

Jonathan,

I took some time out this afternoon to examine the image you offered and make some tests, which I actually printed, taking on board Jeff's advice that one does need to look at the printed outcome when it comes to sharpening.

I notice that the version you posted is at a resolution of 100ppi. That really is below the comfort level for such testing, but it works. The more usual situation would be higher resolution images, so what I came up with may be of limited general validity, but nonetheless here goes.

The first thing I did was a careful examination for noise. I do this with Noiseware Professional. I converted a copy ofthe image to Lab to insure I was getting a clear analytical separation between luminance and colour noise. At all four frequencies it measures, Noiseware found insignificant amounts of colour noise. Only at High frequency was there more than 10% noise and this is confined to the very darkest areas of the image (luminance below level 25, such as the sides of the roof dormers). At most levels the amount of luminance noise ranges from less than 10% to insignificant. The facades of the buildings, the road and the roofs have either insignificant or very low levels of noise - nothing that would bother in a print. So I decided that noise is a non-issue for this image.

Moving on from noise, I turned to PKS and found the optimal settings to be for Capture Sharpen - Hi-Res Digital Superfine Edge Sharpen with opacity of the master layer reduced to 70%, and for Output Sharpen, using Epson Enhanced Matte paper and dimensioning the image (without resampling) to 6*9 inches for printing on a letter-size sheet it comes to 133ppi, therefore the only feasible setting is Matte 180ppi. I reduced the opacity of the master layer for this set to 50%.

Between capture sharpen and output sharpen, I made a Curve Adjustment Layer with level 255 remapped to level 230and input level 24 remapped Output 25. This brightened, increased contrast and re-darkened a bit. Gives the scene more snap with no clipping.

Then I made two versions of this structure: one using John Paul Caponigro's High Pass mid-tone contrast enhancement technique (available on his website) with default opacity of 20% and one without it. This is done with a Stamp Layer and done before Output Sharpening. Comparing these two prints, I preferred the one without the mid-tone contrast enhancement. Combined with Output Sharpening, it's a bit too brittle for my taste.

Then I turned to Focus Magic. I have an evaluation copy of this program, so I had to convert your image to JPEG. That may introduce an impurity in the comparison, so I need to mention this here.

Anyhow, I made the same Curve adjustment layer first in hi-bit mode, did NOT do the local contrast enhancement, flattened, converted to 8-bit, converted to JPEG at highest available quality and Saved-As.

Then I opened this image in FM, and selected the "Focus" filter because for this image none of the others seemed to apply. I thought the resolultion filter would do as well, but it seemed quite ineffective - perhaps because this is an evaluation copy the advertised preview of the effect doesn't show. Since this image has no motion blur, Focus seemed the next most obvious option, based on what is described in their Help manual.

So I made two images with that - one at level 2 and the other at level 4. I printed those as well. The image at level 4 is a toss-away - very obvious halos. The image at level 2 is acceptable. When I put this image side-by-side with the PKS image excluding the mid-tone contrast enhancement, I find the PKS result generally preferable. It is "sharper" (but perhaps had I run FM at level 3 it would have been awash!), the detail of texture in the road and the roofs is better - and this is not sharpening noise, I'm talking about real edges. But more a propos, the FM result seemed uneven - it did a pretty good job on the left hand-side of the image, and as one moves to the right, the quality falls off relative to the PKS image - such that the blue and white sign for the M&S shop is very obviously more distinct in the PKS result than in the FM result.

That's as far as I've taken it.

Cheers,

Mark