deconvolution sharpening plug in

[Robert Ardill]

Hi Tim ... could you make your original raw file available to play around with?  I would like to see what the result is like with InFocus etc.

Robert

[Hening Bettermann]

Hi Tim, thank you for the demo. I note that you use Linear Light as the blend mode, rather than Luminosity; and that in your base layer, you have defined 2 overlapping areas. It looks like, amongst others, I need some basic read-up on layer blending, with wich I am not familiar at all.
- As for the look, I prefer the middle image, without the highpass sharpening. The one at right has a little grainy look to my eyes.

[earlybird]

I think I found out why you are advised to use piccure+ before cropping an image. Piccure+ seems to work well If you choose to "Delete cropped pixels" when cropping an image in Photoshop.

If you elect to not "Delete cropped pixels" when cropping and later launch piccure+ it crashes before it successfully opens to the plug in interface.

I am guessing it just seems easier to tell people to crop after rather than explain the details of the circumstance.

[earlybird]

Finally got around to trying Topaz InFocus. It seems to really dig in and bring out the detail. Perhaps it is too much, too early, if used as a pre sharpener? I look forward to having 30 days of trial and error with it. :-)

[Bart_van_der_Wolf]

Finally got around to trying Topaz InFocus. It seems to really dig in and bring out the detail. Perhaps it is too much, too early, if used as a pre sharpener? I look forward to having 30 days of trial and error with it. :-)

Hi,

It's easy to overdo the deconvolution with Topaz InFocus, which would create artifacts. As a tip on how to avoid that, see my earlier post in this thread.

Cheers,
Bart

[Robert Ardill]

I've just downloaded the piccure+ free trial and I'm getting a very ugly result straight off.  Here is a crop of a slanted edge in piccure+ with all settings off.  you can see the white line inside the black box very clearly, and also the halo around the box.

The photo is with all sharpening off in Lightroom. The image looks fine in Photoshop (but not after piccure obviously).

You can try this by putting a dark gray box on a light gray background and then running piccure.

Either my setup is faulty in some way (don't think so) or piccure+ is for the recycle bin!

[earlybird]

Hi,

It's easy to overdo the deconvolution with Topaz InFocus, which would create artifacts. As a tip on how to avoid that, see my earlier post in this thread.

Cheers,
Bart

Thank you Bart, your advice is very similar to the way I was using it this morning. It seems so *powerful* that I need to teach myself what the threshold of overdoing it is. :-)

[earlybird]

I've just downloaded the piccure+ free trial and I'm getting a very ugly result straight off...
Either my setup is faulty in some way (don't think so) or piccure+ is for the recycle bin!

Hi Robert,
 I think your test example is a worst case candidate for showing piccure+'s ability. I have been getting bad results with high contrast edges being output with wide dark lines. Piccure+ seems to work very well on frames filled with lots of lower contrast and soft detail. Even when I find that piccure+ is helpful I find that the range of parameter adjustments seems crude. Sometimes I make a one click change in a single parameter and can barely, if at all, see a change in output while other times I make a single parameter change and the results are suddenly ugly. In other words, adjusting the parameter sliders does not seem to result in subtle differences that can be appreciated.

 I suspect I will be purchasing a license for both piccure+ and Topaz InFocus and continuing to learn when one is more helpful than the other.

 When I consider how piccure+'s display doesn't show correct color with ProPhotoRGB and costs so much more than InFocus it seems a bit overpriced. I can buy a InFocus license to share on mine and my wife's computer for $52 but piccure+ costs $119.

 My general impression of convolution mathematics is that it does not involve sophisticated programming but rather requires brute force computational power as provided by the hardware. That suggests to me that piccure+ should be spending more time on improving the graphical interface to make it seems like a good value.

[robgo2]

I understood that smart sharpen in PS used deconvolution sharpening, but perhaps I've got that wrong.

In Lightroom (and ACR) the sharpening uses deconvolution if the detail slider is moved right.  No convolution when the detail slider is at the left end, or so Jeff Schewe says in "The Digital Negative".

This is my understanding as well--the Sharpening/Detail slider in ACR/LR uses deconvolution. However, in my experience, it is a weak and poor implementation of the process.

There are other raw converters that utilize deconvolution sharpening much more effectively. Photo Ninja is my particular favorite. DxO's lens blur correction is deconvolution as well. This approach appeals to me, because it allows "optimal" capture sharpening in the raw stage.

Rob

[Robert Ardill]

I think your test example is a worst case candidate for showing piccure+'s ability. I have been getting bad results with high contrast edges being output with wide dark lines. Piccure+ seems to work very well on frames filled with lots of lower contrast and soft detail. Even when I find that piccure+ is helpful I find that the range of parameter adjustments seems crude. Sometimes I make a one click change in a single parameter and can barely, if at all, see a change in output while other times I make a single parameter change and the results are suddenly ugly. In other words, adjusting the parameter sliders does not seem to result in subtle differences that can be appreciated.

Hi Eric,

The slanted edge analysis is a pretty standard way of testing imaging systems at this stage and if a sharpening algorithm shows serious artifacts either visually or using an analysis tool like Imatest then I think it's fair to conclude that the algorithm has major issues.

I've run Focus Magic, InFocus and piccure+ on the slanted edge, taking as much care as I could to err on the low rather than the high side (so, for example, in Focus Magic serious artifacts were obvious at 4 pixel radius but I dialled down to 2 pixels). Here are the results:



InFocus gives the best results by far, except at low frequencies where Focus Magic holds up the contrast better.

Focus Magic gives a good result except at high frequencies where it pretty much loses it completely.

piccure+gives a good edge rise ... however it achieves this by applying a strong dark halo.  The MTF50 result is poor and the MTF-Nyquist is very poor.

It's clear to me that InFocus is the winner - and furthermore it has a far superior interface to FocusMagic.  I wouldn't touch piccure+.  Focus Magic, on the other had, does a very decent job at lower frequencies, is very simple to use and very fast. Further sharpening with a very low radius would probably fix the high frequency softness.


Robert

[earlybird]

Is there a way to use your test method to analyze the results of the various sharpening processes on a photo such as this or do the tests just work on graphical chart type subjects?:

[Tim Lookingbill]

Hi Tim, thank you for the demo. I note that you use Linear Light as the blend mode, rather than Luminosity; and that in your base layer, you have defined 2 overlapping areas. It looks like, amongst others, I need some basic read-up on layer blending, with wich I am not familiar at all.
- As for the look, I prefer the middle image, without the highpass sharpening. The one at right has a little grainy look to my eyes.

Hi Hening, the grainy look isn't going to standout in a 36x24in. print unless the viewer has their nose right up on it which is what that zoom level at 100% in Photoshop shows.

My experiment was to show how blending the bottom layer's shadow into highlight ratio is more controllable than Smart Sharpen and ACR/LR's masking slider. In fact the posted shadow slider positions are blend optimized to take advantage of upsampling's softening of noise with similar effect to Luminance noise reduction in ACR/LR but with a more controllable masking effect. I really don't like ACR/LR's masking slider. I never use it. I just back off Detail slider, but this is in CS5-ACR and LR4.

Robert, I'm not into uploading my Raw files. It's a PITA and besides you can find dozens of Raw shots of duck images to download online to conduct sharpening experiments. I mean how hard is it for you to go to your local park and take a shot of ducks?

[Robert Ardill]

Is there a way to use your test method to analyze the results of the various sharpening processes on a photo such as this or do the tests just work on graphical chart type subjects?:

No, unfortunately not.  The slanted-edge analysis can only be used on a test image.  I think it's useful for comparing different sharpening methods and it can show the effect of over-sharpening and under-sharpening, poor contrast, low resolution, different lenses, f-stops etc., but that's it.  These findings can then be applied to real-life images, but the images can't be analysed as a slanted edge can.

Robert

[Robert Ardill]

Robert, I'm not into uploading my Raw files. It's a PITA and besides you can find dozens of Raw shots of duck images to download online to conduct sharpening experiments. I mean how hard is it for you to go to your local park and take a shot of ducks?

True, true ... but then it wouldn't be YOUR duck now, would it Tim?  But never mind, I do have lots of pics of ducks and herons and swans and egrets and so on and so forth.  Here's one that I sharpened with InFocus and Detail.  As you can see the original image was rather soft, having been taken in low light at 200mm, f11 and 1/60th of a second, hand-held ... not to mention the swan's movement.  My hand probably shook in the same direction as the swan's motion

[joofa]

Is there a way to use your test method to analyze the results of the various sharpening processes on a photo such as this or do the tests just work on graphical chart type subjects?:

No, unfortunately not.  The slanted-edge analysis can only be used on a test image.  ... but the images can't be analysed as a slanted edge can.

Huh. Of course, you have been living in the olden times of, what's the name again, yeah, slanted edge method ... The world has moved on to using JIDM on real images:

[Robert Ardill]

No, unfortunately not.  The slanted-edge analysis can only be used on a test image.  ... but the images can't be analysed as a slanted edge can.


Huh. Of course, you have been living in the olden times of, what's the name again, yeah, slanted edge method ... The world has moved on to using JIDM on real images:

What does a value of 0 to 1 tell you about the performance of an imaging system?  That it has more or less resolution? More or less acutance? More or less of both?

A slanted-edge analysis can give you a huge amount of detailed information about the whole imaging system, from lens-style MTF (which shows the performance of the imaging system across the whole frame), chromatic aberration, edge profile, noise analysis ... and other things like distortion, tonal response and color fidelity with different charts.  It also provides a standardized way of comparing different camera/lens combinations, different sharpening or resizing methods, the effect of lens profile correction on resolution etc.

Not that I'm an expert so perhaps I don't understand what JIDM can do?


Robert

[Bart_van_der_Wolf]

What does a value of 0 to 1 tell you about the performance of an imaging system?  That it has more or less resolution? More or less acutance? More or less of both?

Correct, as in most single number qualifiers, they only mean that there is a difference. How significant that difference is, is anyone's guess.

A slanted-edge analysis can give you a huge amount of detailed information about the whole imaging system, from lens-style MTF (which shows the performance of the imaging system across the whole frame), chromatic aberration, edge profile, noise analysis ... and other things like distortion, tonal response and color fidelity with different charts.  It also provides a standardized way of comparing different camera/lens combinations, different sharpening or resizing methods, the effect of lens profile correction on resolution etc.

And that's why it is an ISO approved method for measuring Resolution for digital scanners and cameras. There is a lot of information that can be extracted from the results. One of the things it clearly demonstrated in the examples you showed, is that FocusMagic is very good at restoring detail below the Nyquist frequency and it avoids creating aliasing artifacts, and Infocus also boosts signals (above the Nyquist frequency) that may lead to aliasing but can also look sharper at the limiting resolution.

Implementations of the ISO procedure like Imatest does, also allows to view the data at a number of ways, highlighting different aspects of the results. It is also one of the few methods that allows to study the behavior at higher spatial frequencies than the Nyquist limit, because the slanted edge allows to super-sample the pixels at 4x the Nyquist frequency (it's actually sampling at close to 10x, for a 5-6 degree slant, but for statistical robustness it bins the results in larger bins).

Cheers,
Bart

[Bart_van_der_Wolf]

Is there a way to use your test method to analyze the results of the various sharpening processes on a photo such as this or do the tests just work on graphical chart type subjects?

Hi,

The difficulty with analyzing a random/real image, rather than one shot under strictly normalized conditions, is that there are numerous variables that cannot be individually identified. It's like trying to unscramble an omelet. Is the blur caused by subject motion, lens aberrations, camera shake, defocus, diffraction? How much of these factors is present in the image with what weigthing, and how much is signal and how much is noise, or in case of JPEGs added lossy compression artifacts?

There are some measurements possible, e.g. how much of the various spatial frequencies are present in an image by means of a periodogram, but it is hard to compare results between images from different subjects, it is only useful to quantify content from a particular image. So it would only work on identical images with a before/after comparison, and still needs interpretation based on subject matter and how it was captured.

Cheers,
Bart

[Jack Hogan]

I've just downloaded the piccure+ free trial and I'm getting a very ugly result straight off.

Robert,

As mentioned earlier, piccure+ themselves suggest that there are better tools for capture sharpening images taken at their sharpest f-numbers with good technique.  They on the other hand claim to be able to perform blind reversal of the effects of spatially-varying optical aberrations.  This is one mean feat and requires major computational power.  But unless you have such a situation use the other simpler, faster tools.

Incidentally, one of InFocus' neatest features is its one click capture sharpening.  To use it zero out the Sharpen section and set up the following as a preset, it comes straight from dr. Albert Yang, President of Topaz:

Blur Type: Unknown/Estimate
Blur Radius: 2 (don't worry, it does not mean 2 pixels in this context)
Edge Softness: 0.3

The next time you want to capture sharpen an image bring it into InFocus, recall the preset and click the 'Estimate Blur' button.  Works pretty decently most of the time.
Jack

[Robert Ardill]

And that's why it is an ISO approved method for measuring Resolution for digital scanners and cameras. There is a lot of information that can be extracted from the results. One of the things it clearly demonstrated in the examples you showed, is that FocusMagic is very good at restoring detail below the Nyquist frequency and it avoids creating aliasing artifacts, and Infocus also boosts signals (above the Nyquist frequency) that may lead to aliasing but can also look sharper at the limiting resolution.

Yes, I think it's clear that FM does a really great job up to MTF40 as can be seen below.  The MTF80 result is way better than the InFocus result.  So the FM image should look sharper overall.  InFocus pulls up the MTF near Nyquist, but that seems to be caused by aliasing.  I sharpened the InFocus result a second time with a small radius and the jaggies are quite obvious at 300% as can be seem in the image under the graphs.

The one thing I've found though is that it's really necessary to dial-down the FM setting quite a bit.  For example with this image the artifacts became strong with a radius of 5, but I had to drop the radius down to 2 in order to get a clean edge profile and MTF.  Even at 3 there is significant overshoot.

So my take right now (I seem to be doing a bit of a two-step here ) is that FM is the better of the two, but it's really important to step the radius back by at least 2 from the artifact radius.  Otherwise corrective action (like the layer blending you suggested or reducing the sharpened layer opacity) will be necessary to get rid of the halos.



So a possible approach would be to use Focus Magic with a dialed-down radius, followed by some very low radius unsharp-mask sharpening, giving the sort of result below:




Robert