deconvolution sharpening plug in

[Robert Ardill]

I have also downloaded and installed a Focus Magic demo for the second time in 12 months and I was reminded that the results I get with it seem, to put it politely, *less than good*.  I am surprised that so many here seem to find it useful, so I guess I will keep trying it on occasions to see where it works well.

The trick for me, with Focus Magic, is to up the radius until you see halos/artifacts, and then to step back at least one pixel and usually better 2 pixels. Just like with all these other techniques, if you use too high a radius you will get halos/artifacts.

An alternative which I also think is very good is to step back only 1px and then do a Fade.  That gives very good control over any small amounts of haloing that might be present.

Robert 

[Bart_van_der_Wolf]

The trick for me, with Focus Magic, is to up the radius until you see halos/artifacts, and then to step back at least one pixel and usually better 2 pixels. Just like with all these other techniques, if you use too high a radius you will get halos/artifacts.

Yes, I use a similar technique, but I also start with an amount setting of 300% to exaggerate the effect for easier detection of when the artifacts become visible. I then step back 1 blur width from the sudden creation of fatter features and contours, and then adjust the amount to more reasonable values. Strictly speaking, halos should not happen with deconvolution Capture sharpening with the proper blur width setting. When used for Creative sharpening one can increase the amount, not the blur width.

An alternative which I also think is very good is to step back only 1px and then do a Fade.  That gives very good control over any small amounts of haloing that might be present.

I use a blend-if layer setting (see attached) that avoids clipping, and that mitigates the restoration in regions that already have a high edge contrast (which allows to use a somewhat higher amount).

After many years of comparing alternatives, I find FocusMagic to be one of the best at improving the signal to noise ratio, i.e. it doesn't sharpen noise as much as it does the signal, and it generates very few artifacts. People who need more than real resolution restoration (i.e. looking for an effect that suggests sharpness), should additionally consider using Topaz Detail (it performs miracles for the rendering of structural detail).

Cheers,
Bart

[Robert Ardill]

Yes, I use a similar technique, but I also start with an amount setting of 300% to exaggerate the effect for easier detection of when the artifacts become visible.

Good thinking.

Strictly speaking, halos should not happen with deconvolution Capture sharpening with the proper blur width setting. When used for Creative sharpening one can increase the amount, not the blur width.

Yes, that is quite clear either visually or with the slanted edge analysis.  The reason I use Fade in Photoshop rather than reduce the Amount in Focus Magic is that the preview in FM is quite poor and so it's much easier to see the effect in Photoshop.  But are the two equivalent??  Perhaps you could answer that question ... but I'll try it out with Imatest also, from an experimental rather than theoretical point of view.

I use a blend-if layer setting (see attached) that avoids clipping, and that mitigates the restoration in regions that already have a high edge contrast (which allows to use a somewhat higher amount).

Yes, I do that too ... especially if I apply a stronger creative-type sharpening which results in some haloing (even if it isn't supposed to ).

After many years of comparing alternatives, I find FocusMagic to be one of the best at improving the signal to noise ratio, i.e. it doesn't sharpen noise as much as it does the signal, and it generates very few artifacts. People who need more than real resolution restoration (i.e. looking for an effect that suggests sharpness), should additionally consider using Topaz Detail (it performs miracles for the rendering of structural detail).

Topaz Detail is great ... I agree. What about InFocus?  Not as good as FM in your view?

[Bart_van_der_Wolf]

Topaz Detail is great ... I agree. What about InFocus?  Not as good as FM in your view?

Topas Infocus is good, but it's more prone to generating artifacts than FocusMagic, so it needs the exactly correct settings for especially the Blur radius to be used. Also in this case, I start with the additional Sharpen panel (Micro Contrast and Sharpness) settings maxed out to 1.00 and something like a 0.80 Sharpness radius, to improve/exaggerate the visibility of deconvolution artifacts showing up as one increases the Blur radius value.

I usually start around 0.5 Blur radius, without any artifact suppression (Suppress Artifacts slider = 0.00). Then, once clicked I can increase the Blur radius slider values with the arrow keys of my keyboard in very small amounts upwards until the exaggerated preview starts showing (mostly ringing) artifacts. I then reduce the first visibility of that with the Suppress artifacts slider, and then bring back the additional Sharpen panel's settings to realistic values. Since the Deblur panel only serves as a Capture sharpening set of controls, the Sharpen panel needs to be used to build on that improved resolution foundation to add some amplitude.

Don't make the common mistake to attempt increasing the amount of sharpening with the wrong control, i.e. the Blur radius control. It's not made for that, it only restores natural sharpness. It's the Sharpen panel's duty to control the amount, and modify a bit of the shape (the spatial frequencies that get boosted) with its radius.

Proper deconvolution sharpening mainly restores micro-contrast resolution, i.e. at the limiting spatial frequencies. That will lift the entire MTF curve, without overshoots. Micro-contrast is the contrast between the pixels with the highest spatial frequencies, the micro detail.

One can boost other/lower spatial frequencies with other tools, usually based on wavelet decomposition, but that's basically (only much more intelligently) what Topaz Detail allows to do, halo free and color preserving.

Cheers,
Bart

[Robert Ardill]

Topas Infocus is good, but it's more prone to generating artifacts than FocusMagic, so it needs the exactly correct settings for especially the Blur radius to be used.

I've done a quick test with InFocus (using Bart's methodology) and also comparing it to FM and ACR+FM, results below:



As you can see, my first attempt with ACR25 & InFocus with a 1.3 blur radius and best settings for artifact removal (which didn't LOOK too bad) is a disaster (first line), way oversharpened.

On the 3rd line I have InFocus with a 1.3 blur radius again and nothing else. It seemed to me that applying sharpening only really added a halo and microcontrast only seemed to boost noise. At any rate the results are not bad, although the ACR25+FM1 above is better at everything except MTF50.

I then applied Topaz Detail3 (to the Infocus 1p3) at Microcontrast=0,Medium=0.2 and Large=0.2 and that wins on every point against the other three.  Here is the slanted edge and MTF graphs, which look pretty good for both contrast and resolution with essentially no artifacts and an impressive MTF of 2552 lw/ph at Nyquist:

[Robert Ardill]

As a further step to the InFocus+Detail sharpening above, I applied sharpening using ACR at 0.5 radius, quite high value (around 70), Detail set to 0 and Masking to 100 to see if I could sharpen the high frequency components and lift the MTF curve.

The result is very good, I think:



It has tightened up the edge profile to just 1px for the 10-90% rise, which is excellent. The MTF50 has gone to 3973 lw/ph which is outstanding, considering that the camera is an A6000 with 4000 px picture height. And what is even more outstanding is that the MTF at Nyquist is 3920 lw/ph.  There are essentially no artifacts added by the final sharpening.

Bart mentioned that some sharpening needs to be added in InFocus to add amplitude and I think that this is essentially what I have done here ... so the deblur and sharpening can be done in InFocus. However ACR has more control over the sharpening as it has a Detail slider and Mask so that sharpening artifacts can be minimised.

Here is a real-life comparison (100% crop) using the same settings exactly (top has no sharpening, bottom has InFocus+Detail+ACR:

[Tim Lookingbill]

Robert, the bottom sharpened result looks artificial. It looks too overly smooth and clay like.

Or are you sharpening so it looks natural when downsized for web viewing which is output sharpening and not capture sharpening.

[Robert Ardill]

Robert, the bottom sharpened result looks artificial. It looks too overly smooth and clay like.

Or are you sharpening so it looks natural when downsized for web viewing which is output sharpening and not capture sharpening.

Hi Tim ... well they both look smooth and clay-like to me because it is a clay-like render over pebble (pebble-dash https://en.wikipedia.org/wiki/Roughcast).  The reason I picked that photo is a) because I had it to hand, and b) because the original is quite soft as you can see. 

I didn't downsize the image for web viewing or output sharpen it ... this is just capture sharpening with a crop at 100%, captured using MWSnap off my monitor.  It could be that MWSnap isn't doing too good a job.

Here is a crop of another image that I took this evening behind my house with my brand new Batis 85mm .  You really need to download (right-click) and view it in Photoshop.



I don't think the image really needed any sharpening (none was applied in Lighroom) and if I was using it I would dial-down the sharpening, but this is for illustration. The fine detail has been brought out and there is stronger contrast (this is from Topaz Detail).  Noise has been boosted a bit, but you need to go to 200% to see that.  I suppose an edge mask could be used  for noisier images to avoid boosting noise.

Robert

[Tim Lookingbill]

So what was your point behind the painted wall sample? What do you want us to understand about sharpening outside of using visual judgement vs nyquist graphs?

It appears from the painted wall image that there's not a lot of acuity in the lens you're using or you're over exposing and introducing flair or nonlinear sensor behavior.

I get a lot of Raw shots similarly with a $500 DSLR/kit lens system, but I know I can make a silk purse out of a sow's ear by remapping the tonality of the nonlinear/flat looking image and then apply the right amount of sharpness that fits hand in glove with the tonality as I've demonstrated in the sample below and that's a jpeg.

I pretty much have to do this on every image I shoot more or less. Scientific analysis isn't going to tell you anything about making a better looking image on a consistent basis. Every scene/exposure combination is going to introduce some kind of nonlinear behavior that has to be fixed in post.

[Tim Lookingbill]

Thanks, Bart, on that layer blend tip. It gave me an idea about applying it to a high pass sharpening routine on severely upsampled low resolution images with better results than using Smart Sharpen in CS5.

[Hening Bettermann]

Concerning that same layer arrangement: a dumb question: which are the 2 layers you are blending? 2 dups of the background, sharpened in 2 different ways/degrees?

[Tim Lookingbill]

Are you asking Bart or me?

[Bart_van_der_Wolf]

Concerning that same layer arrangement: a dumb question: which are the 2 layers you are blending?

The screen copy in reply 81 shows the effect of regular (small radius) sharpening, which clips, and the same sharpened image with a Blend-if on top of the unsharpened original.

So the procedure would be to make a duplicate layer of e.g. the Background layer (or a merge of the visible layers), sharpen that, and use a Blend-if Luminosity layer to gradually blend it with the original as the edge contrast approaches clipping. The idea behind it is that if edge contrast is already high, it already looks sharp, and it's better to avoid clipping artifacts instead of exaggerating the sharpening.

Cheers,
Bart

[Bart_van_der_Wolf]

Thanks, Bart, on that layer blend tip. It gave me an idea about applying it to a high pass sharpening routine on severely upsampled low resolution images with better results than using Smart Sharpen in CS5.

You're welcome Tim. I use a Photoshop action that duplicates a layer, sets the Blend-if parameters and Luminosity blending, and then calls (in my case) FocusMagic, but it could call any filter or action that sharpens. That makes it a 1 mouseclick operation, if the Blend-if parameters are acceptable, and whatever sharpening radius or other parameters are needed. The layer can be switched off or removed, e.g. before resampling, or it can be partially masked where the sharpening is not needed. It's very flexible.

Cheers,
Bart

[Robert Ardill]

So what was your point behind the painted wall sample? What do you want us to understand about sharpening outside of using visual judgement vs nyquist graphs?

It appears from the painted wall image that there's not a lot of acuity in the lens you're using or you're over exposing and introducing flair or nonlinear sensor behavior.

Hi Tim,

I just had a look at the image and you're right ... what I posted shows nothing of interest.  It's because of the way I captured it.  Here is crop, as before, but this time straight out of Photoshop.

The lens I used is a superb lens (Sony/Zeiss 55mm f/1.8 ) but the focusing was not 100% as you can see.  What I am trying to show in the images is the same as in the the MTF chart, that fine detail is restored as is contrast, with no ugly artifacts like halos.  Have a try on the original image if you like: A6000-55mm-Wall



I guess this is a question of whether of not one is willing to trust the result of the sort of experiment that does not rely on a subjective evaluation.  The MTF/Edge profile analysis is intended to show the performance of an imaging system either with or without post-processing.  I personally think that it is very useful as a tool to evaluate different sharpening methods (or resizing methods) ... but of course that's a personal opinion. If you feel that it is better to make a judgement based solely on real-life images then that's fine; but I find that it's really quite easy to think that one is sharpening well when in fact we are damaging the image.  Perhaps that doesn't matter in most instances if the result is pleasing, but when we are trying, for example, to make large prints for close-up viewing as in a gallery ... well then every little bit that we can squeeze out of the photo is worth doing.

[Tim Lookingbill]

Robert, that second attempt looks more real (natural) which is all that I look for.

But it still looks overly smooth but at least more consistent overall which masking tends to introduce unevenness between smooth surfaces next to fine crispy detail. The eye sees fine detail like granules from dirt and tiny dried bubbles and cracks underneath the paint along smooth raised bumps. But your second attempt makes me think the viscosity of the paint must've been thicker and dried to a burnished look differently across the entire surface.

It just looks odd in a surreal way, but maybe that's what will make the print unique from an aesthetic standpoint. But scientific analysis is not going to predict that kind of outcome on a consistent basis.

I couldn't download your ARW file. I get a "Enable Quicktime" alert in Firefox and when I allow it, I end up with a blank white page.

[Robert Ardill]

Robert, that second attempt looks more real (natural) which is all that I look for.

But it still looks overly smooth but at least more consistent overall which masking tends to introduce unevenness between smooth surfaces next to fine crispy detail. The eye sees fine detail like granules from dirt and tiny dried bubbles and cracks underneath the paint along smooth raised bumps. But your second attempt makes me think the viscosity of the paint must've been thicker and dried to a burnished look differently across the entire surface.

It just looks odd in a surreal way, but maybe that's what will make the print unique from an aesthetic standpoint. But scientific analysis is not going to predict that kind of outcome on a consistent basis.

I'm not using masks here Tim ... I just mentioned that with very noisy images that an edge mask could be used to reduce the noise-boosting.

The photo certainly isn't meant to be artistic . I took it to test my camera/lens because the variations in texture and detail is quite useful (and it also happens to be the back wall of my house).

I couldn't download your ARW file. I get a "Enable Quicktime" alert in Firefox and when I allow it, I end up with a blank white page.

Try Chrome ... the download works for me.

[Hening Bettermann]

Are you asking Bart or me?

Everybody who knows and is willing to explain ;-)

[Hening Bettermann]

Bart, thank you for the explanation of the layer blending!

[Tim Lookingbill]

Everybody who knows and is willing to explain ;-)

Just experimenting capture sharpening for uprezing in LR4 to 36in x 24in/240ppi my 6MP Raw cropped to 4MP and doing further edge mask sharpening in CS5 Photoshop using Highpass/2.50 setting on a Linear Light blend mode layer. See the screengrabs. The actual zoom of the image is how it looked in LR4 at 4:1 zoom view to match the 100% zoom in Photoshop. The feather detail in the middle has the highpass layer turned off and the one on the far right has it on.