Pages: [1]   Go Down

Author Topic: Help needed with understanding deconvolution sharpening  (Read 7605 times)

Ad_Astra

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
Help needed with understanding deconvolution sharpening
« on: January 15, 2017, 01:25:47 pm »

Hi

I am looking for some guidance please on how to use deconvolution sharpening as a replacement for capture sharpening in Lightroom or ACR.

I am mostly OK with Photoshop techniques and I have read various threads on deconvolution sharpening but I am puzzled by the following and would value any insight people can share.

1) When to use a plugin such as Focus Magic in the workflow.

I understand that the capture sharpening needs to be done early in the workflow. I also have read that things like luminance noise reduction and vertical upright correction should be done after deconvolution sharpening. Normally I would do all these in Lightroom. If I need a round trip to Photoshop to run a plug-in what is the recommended workflow to tackle ACR corrections, deconvolution sharpening, luminance noise reduction and converging vertical corrections?

2) Preparation steps

I have read that it is better to upscale an image and convert to a linear gamma before running a deconvolution plug-in. What is the recommended way to do these tasks? I have no idea on how to convert to linear gamma and back to pro-photo gamma 1.8.

3) Which plug-in to use?

I have the Topaz suite so have Topaz InFocus. I am happy to consider an alternative plug-in. What is the preferred plug-in for carrying out capture deconvolution sharpening? I have heard of Focus Magic and Piccure +, are these worth the extra cash compared to InFocus?


Many thanks



Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Help needed with understanding deconvolution sharpening
« Reply #1 on: January 16, 2017, 06:49:00 am »

Hi

I am looking for some guidance please on how to use deconvolution sharpening as a replacement for capture sharpening in Lightroom or ACR.

I am mostly OK with Photoshop techniques and I have read various threads on deconvolution sharpening but I am puzzled by the following and would value any insight people can share.

1) When to use a plugin such as Focus Magic in the workflow.

I understand that the capture sharpening needs to be done early in the workflow. I also have read that things like luminance noise reduction and vertical upright correction should be done after deconvolution sharpening. Normally I would do all these in Lightroom. If I need a round trip to Photoshop to run a plug-in what is the recommended workflow to tackle ACR corrections, deconvolution sharpening, luminance noise reduction and converging vertical corrections?

Hi,

Yes deconvolution is best done early in the workflow, because it avoids additional alterations of the original source data. To better understand that, it's pehaps a good idea to recap what deconvolution does.

Deconvolution is a mathematical attempt to reverse a so-called convolution operation. A convolution is a mathematical description of how the (in our case) energy of point light source or a point from the surface of our subjects is also spread over several neighboring pixels instead of a single one. The model used to describe that convolution is called a Point Spread Function (PSF). Using the closest approximation of that PSF model, a deconvolution pulls that spread out energy back to the source position by subtracting it from the surrounding pixels and adding it to the central pixel. This is then repeated for all pixel positions.

A significant problem is that the signal levels in the central pixel under investigation, and the neighboring pixels, do not only come from the original subject's brightness at a given exposure time but also contain a lot of noise. In fact, light itself is noisy due to the random arrival times of the photons, and the camera electronics also add (a different type of) noise into the mix. That makes it harder to determine exactly how much signal energy to get back from neighboring pixels while leaving the noise mostly unaffected.

So, any alteration of the original pixels will change the somewhat predictable noise distributions and make it harder to only deconvolve the signal and not the noise that was inherent or added to that pixel but not blurred. Noise reduction is often not a linear operation but it adapts to the image contents and attempts to preserve detail. Distortion corrections need to resample the image data, and resampling introduces its own blur (in a non uniform way!). This all needlessly complicates, or even prevents, successful deconvolution.

Even tone-curve adjustments or a gamma curve complicates the restoration of the original signal component, because it disturbs the RGB balance. That's why a linear gamma data set is preferred for accuracy and allows to use relatively simpler calculations.

Quote
2) Preparation steps

I have read that it is better to upscale an image and convert to a linear gamma before running a deconvolution plug-in. What is the recommended way to do these tasks? I have no idea on how to convert to linear gamma and back to pro-photo gamma 1.8.

For the better quality deconvolution plugins, that's usually not necessary. Who knows, they may already do some of that under-the-hood. Resampling only the scale of the image will usually introduce artifacts of its own, and we do not want to exacerbate those. However, if done with care, there might be some very slight benefits to being able to use a more accurate PSF model that can improve sub-pixel accuracy. But for most common situations, I'd let the plugins do their thing of the as-original-as-possible image data, which is hard enough given the amount of tonal adjustments that are applied by default during Raw conversion.

Quote
3) Which plug-in to use?

I have the Topaz suite so have Topaz InFocus. I am happy to consider an alternative plug-in. What is the preferred plug-in for carrying out capture deconvolution sharpening? I have heard of Focus Magic and Piccure +, are these worth the extra cash compared to InFocus?

Which plugin to use, is a bit of a moving target as new versions are introduced, and with per image effectiveness. But given that you already have Topaz Labs InFocus, you can already do a lot to improve the resolution of your images, even with the drawback of partially altered image data.

The benefit of FocusMagic is that it's somewhat less prone to generating artifacts and it makes a good separation between signal and noise, but with proper care, they can also be mostly avoided with 'InFocus'. Piccure+ is a bit of a different animal, but I'm not confident that it doesn't also reduce image gamut. How Color Management is affected by Piccure+ is not clear to me.

To get the most from your Topaz InFocus, I recommend trying the 'Unknown/Estimate' blur type with a blur radius of 2. That will often get you in the ball park, if you make sure to first zoom in on the sharpest part of the image.

If you want to use the 'Generic' or 'Out of Focus' blur types, it can help to find the proper Radius setting by first maxing-out the Micro Contrast and Sharpness sliders with a Sharpness radius set to 0.8. This will grossly exaggerate the sharpness of the deconvolution, but it will make it easier to see when artifacts due to too large a 'Blur radius' setting are created. I'd start with a 'Blur radius' of 0.5 and slowly increase it until halos and ringing artifacts start to appear, then back off a bit and further mitigate those remaining artifacts with the 'Suppress Artifacts' control. Then set the "Sharpen" amounts back to zero and slowly increase those to your liking.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

scyth

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 584
Re: Help needed with understanding deconvolution sharpening
« Reply #2 on: January 16, 2017, 10:15:48 am »

So, any alteration of the original pixels will change the somewhat predictable noise distributions

old comment from Eric Chan about software optics correction (in that case - ACR/LR interpreting optics correction tags embedded in Panasonic' RW2 raw files from m43 camera/lens) :

"The bands are introduced because uniform image noise becomes non-uniformly distributed upon image resampling (e.g., when spatially-varying optical corrections are applied). Ideally, one would redistribute the image scene values to make the rendered image rectilinear, without redistributing the noise (thereby keeping it uniform in appearance). Regrettably, that is not easy to do, since the noise is already "burned into" the image."

and like it or not - with ACR/LR and raws from certain camera manufacturers you simply can't switch embedded optics corrections off (unless you modify the raw yourself before ACR/LR =  a documented way will be to dump to DNG and correct documented DNG tags before feeding to ACR/LR)...
« Last Edit: January 16, 2017, 10:19:02 am by scyth »
Logged

Ad_Astra

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
Re: Help needed with understanding deconvolution sharpening
« Reply #3 on: January 17, 2017, 05:35:58 pm »

Many thanks Bart for your detailed reply. The description of the Point Spread Function was particularly helpful as on my first pass reading on the threads in this forum PSF did not mean much to me, now it is much clearer.

I think my workflow could use Topaz InFocus "early on" as I have a denoise plugin for Photoshop that I haven't used for ages since ACR added a decent denoise tool. I also have DxO ViewPoint which I also have stopped using since Adobe added the Upright features. I could start using these again after InFocus.

I have a couple of follow up questions if I may please.

1) Could deconvolution sharpening help with a depth of field issue?

I was hoping to rescue a coastline rockface photo suffering from a blurred foreground due to lack of depth of field and limit the effect with a layer mask but my initial attempts haven't been that successful.

2) How does the detail slider in Lightroom and Adobe Camera Raw work? I know from 1 to 50 it applies a halo suppression to the Unsharp mask sharpening. Above 50 it applies a kind of deconvolution sharpening. My question is this deconvolution sharpening on top of the USM sharpening or a replacement for it? E.g. at 100 it is all deconvolution sharpening and no USM or a full USM and full deconvolution sharpening?

I find that I have to keep the detail slider at well below 50 to prevent artefacts so have not explored using values above 50 in much detail.

Also do you have a web site with details of your work? I am sure it would be interesting read if you maintain such a site.
Logged

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 11311
    • Echophoto
Re: Help needed with understanding deconvolution sharpening
« Reply #4 on: January 17, 2017, 07:32:28 pm »

Quote
1) Could deconvolution sharpening help with a depth of field issue?

I was hoping to rescue a coastline rockface photo suffering from a blurred foreground due to lack of depth of field and limit the effect with a layer mask but my initial attempts haven't been that successful.

Yes and no. Deconvolution can handle a small amount of defocus, but defocus is quiet problematic.

Quote
2) How does the detail slider in Lightroom and Adobe Camera Raw work? I know from 1 to 50 it applies a halo suppression to the Unsharp mask sharpening. Above 50 it applies a kind of deconvolution sharpening. My question is this deconvolution sharpening on top of the USM sharpening or a replacement for it? E.g. at 100 it is all deconvolution sharpening and no USM or a full USM and full deconvolution sharpening?

I find that I have to keep the detail slider at well below 50 to prevent artefacts so have not explored using values above 50 in much detail.

I would say it depends. It is quiet OK to use 100% detail, but you need a small radius, matching the lens. Amount should be say 45 and you would need some noise reduction with masking.

I like to use FocusMagic on a separate layer and mask.

Best regards
Erik


« Last Edit: January 17, 2017, 07:45:32 pm by ErikKaffehr »
Logged
Erik Kaffehr
 

Tony Jay

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2965
Re: Help needed with understanding deconvolution sharpening
« Reply #5 on: January 17, 2017, 07:55:18 pm »

....
2) How does the detail slider in Lightroom and Adobe Camera Raw work? I know from 1 to 50 it applies a halo suppression to the Unsharp mask sharpening. Above 50 it applies a kind of deconvolution sharpening. My question is this deconvolution sharpening on top of the USM sharpening or a replacement for it? E.g. at 100 it is all deconvolution sharpening and no USM or a full USM and full deconvolution sharpening?

I find that I have to keep the detail slider at well below 50 to prevent artefacts so have not explored using values above 50 in much detail.
...
Just to confirm that the halo suppression is least at 50 and maximal at 0 - so move the slider to the left from 50 to get more effect. If you want deconvolution sharpening go above 50 - in this case moving the slider to the right intensifies the effect.

You probably knew this but it isn't stated explicitly.

Tony Jay
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Help needed with understanding deconvolution sharpening
« Reply #6 on: January 18, 2017, 06:42:40 am »

I have a couple of follow up questions if I may please.

1) Could deconvolution sharpening help with a depth of field issue?

I was hoping to rescue a coastline rockface photo suffering from a blurred foreground due to lack of depth of field and limit the effect with a layer mask but my initial attempts haven't been that successful.

In theory, yes. But in practice, it is very difficult, because the PSF changes with distance and we do not have depth or distance information. If we apply a deconvolution to the more blurred areas we can restore some of the source info, but at the same time we'd over-compensate the already sharp(er) regions in the image. So we'd need to mask regions out, based on their level of blur and, ideally, discriminate between blur in front and in behind the (to be in) focus plane.

Quote
2) How does the detail slider in Lightroom and Adobe Camera Raw work? I know from 1 to 50 it applies a halo suppression to the Unsharp mask sharpening. Above 50 it applies a kind of deconvolution sharpening. My question is this deconvolution sharpening on top of the USM sharpening or a replacement for it? E.g. at 100 it is all deconvolution sharpening and no USM or a full USM and full deconvolution sharpening?

As others have answered, it's a gliding scale between USM and (a sort of) Deconvolution. I'm not so sure about halo suppression though, that's what the masking control is for.

Quote
I find that I have to keep the detail slider at well below 50 to prevent artefacts so have not explored using values above 50 in much detail.

The Deconvolution in ACR/Lightroom is prone to creating ugly looking structure/artifacts, so only low amounts (<65) are tolerable most of the time. The deconvolution is also not very effective.

Quote
Also do you have a web site with details of your work? I am sure it would be interesting read if you maintain such a site.

I used to have some technical pages/web-tools, but I need to redo some of them on a different hosting site when I get some more time. They can be located via the Wayback Machine website archive, but they were not created for modern browsers so they may display a bit less well. I'm currently too busy with shooting and post-processing images, and doing some other software development (BTW related to deconvolution and resolution measurements, and artificial Depth of Field). I try to share some useful info and my findings, most of it here on LuLa.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

Ad_Astra

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
Re: Help needed with understanding deconvolution sharpening
« Reply #7 on: January 21, 2017, 03:43:43 pm »

I have been experimenting with another set of images, this time they are a focus staking set with focus point at different postions which gets around the DOF issue I had with my early test photo.

This made me wonder about the correct way to use deconvolution sharpening for capture sharpening. Should I run InFocus on each of the separate images and then focus blend, or focus blend first and run InFocus on the composite focus image?

P.S. Found an interesting comment in the Topaz training videos, they suggest if the blur is greater than 10 pixles then they recommended down sizing the image to get the size of the blur below 10 pixels.


Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Help needed with understanding deconvolution sharpening
« Reply #8 on: January 21, 2017, 03:59:27 pm »

I have been experimenting with another set of images, this time they are a focus staking set with focus point at different postions which gets around the DOF issue I had with my early test photo.

This made me wonder about the correct way to use deconvolution sharpening for capture sharpening. Should I run InFocus on each of the separate images and then focus blend, or focus blend first and run InFocus on the composite focus image?

I'd mildly deconvolve/sharpen the individual images to make it easier for the Focus-stacker to find the best-focused pixels in each stack layer. Since focus-stacking involves resampling/scaling of the individual images to compensate for the magnification differences (to allow registration), the resampled images fused together will stand another round of mild deconvolution to reduce the resampling blur.

Quote
P.S. Found an interesting comment in the Topaz training videos, they suggest if the blur is greater than 10 pixles then they recommended down sizing the image to get the size of the blur below 10 pixels.

Yes, or do multiple runs of the deconvolution. However, there may be too much information lost to allow restoration, in which case downsampling is the only remedy left.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

Hening Bettermann

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 945
    • landshape.net
Re: Help needed with understanding deconvolution sharpening
« Reply #9 on: January 21, 2017, 06:28:46 pm »

>Should I run InFocus on each of the separate images and then focus blend, or focus blend first and run InFocus on the composite focus image?

Just to offer a second opinion:
I put this same question to Stas Yatzenko from Helicon Focus. He advised against deconvolving the source images.  His concern is that artifacts of the deconvolution might interfere with the focus stacking process. - It should be added that he is not a believer in deconvolution, generally.

There are 2 arguments in favor of his view:
1-  Following his advice, I have never experienced that Helicon had trouble finding the best focussed image due to lack of sharpening (it has in the shadows, if the images are not gamma-encoded).
2- It caters for my laziness...

Regardless, this makes me wonder:
Deconvolving early in the workflow sounds theoretically right - however, it kind of presupposes a perfect deconvolution with no artifacts. I wonder how sure we can be of that. Currently, I postpone sharpening and plan to do it after upsampling to print size. Until then - who needs the sharpening if Helicon doesn't?

Good light! - Hening

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Help needed with understanding deconvolution sharpening
« Reply #10 on: January 21, 2017, 09:02:44 pm »

Regardless, this makes me wonder:
Deconvolving early in the workflow sounds theoretically right - however, it kind of presupposes a perfect deconvolution with no artifacts. I wonder how sure we can be of that. Currently, I postpone sharpening and plan to do it after upsampling to print size. Until then - who needs the sharpening if Helicon doesn't?

Hi Hening,

My comment mentioned "mild" deconvolution before stacking. This is to avoid artifacts that might be misconstrued by the stacking process as actual detail. However, from Stas' point of view, I understand his reluctance. In addition, I know that most people think that they need fewer slices for focus stacking than I do, and when too few slices are available, then deconvolving the slices only makes the partial lack of sharpness, going from slice to slice, more obvious.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

jed best

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 398
Re: Help needed with understanding deconvolution sharpening
« Reply #11 on: January 26, 2017, 02:54:13 pm »

Hi Bart,

What about doing only the basic raw demosaicing then use focus magic. After that, go to photoshop and use the ACR filter and finish what was started in Lightroom or C1.

I would be interested in your thoughts.

Cheers

Jed

Logged

Ad_Astra

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
Re: Help needed with understanding deconvolution sharpening
« Reply #12 on: February 05, 2017, 04:42:15 pm »

Hi jedbest

I think your suggestion would help with the point in the pipeline when when the deconvolution sharpening was done. As far as I understand it the ACR filter in Photoshop is not working on the raw file but is working on the layer as if it were a TIFF file. As such the parametric advantages of ACR or Lightroom are lost and you are adding a new pixel layer into the Photoshop document.

I am not an expert but I settled on doing most of the global adjustments in Lightroom except for in the detail panel I leave out the sharpening and luminosity noise reduction. I then use Topaz InFocus in Photoshop with a mild deconvolution setting to do "capture sharpening".

I tried the Focus Magic trial period but found the interface somewhat old fashioned for me.
Logged

Ad_Astra

  • Newbie
  • *
  • Offline Offline
  • Posts: 17
Re: Help needed with understanding deconvolution sharpening
« Reply #13 on: February 05, 2017, 04:44:43 pm »

p.s. Tony Jay updated his thoughts on another forum and no longer believes in ACR/Lightroom that halo supression is least at 50 and maximal at 0. Maybe he can expand in this thread for completeness.
Logged

Hoggy

  • Full Member
  • ***
  • Offline Offline
  • Posts: 207
  • Never take life, or anything in it, too seriously.
Re: Help needed with understanding deconvolution sharpening
« Reply #14 on: February 05, 2017, 10:23:56 pm »

As far as I understand it the ACR filter in Photoshop is not working on the raw file but is working on the layer as if it were a TIFF file. As such the parametric advantages of ACR or Lightroom are lost and you are adding a new pixel layer into the Photoshop document.

Just for clarity, this depends on if one is working in camera raw or not.  If you open a raw file, then you're working in camera raw - where the parametrics are kept and rendered.  But if one is working on a standard layer and chooses the menu item "filter/camera raw filter", then one is no longer working on the raw data.  And hopefully someone will correct me if I'm wrong, but even when using that filter, there is a certain degree of parametrics - in that one can always go back in and edit the filter settings -- it's just that it's not working with the raw data anymore.

...  At least that's my understanding.  But I'm no Photoshop expert - that program's a beast to me.  :)

p.s. Tony Jay updated his thoughts on another forum and no longer believes in ACR/Lightroom that halo supression is least at 50 and maximal at 0. Maybe he can expand in this thread for completeness.

My understanding is that it's a completely gradual sliding scale on the detail slider.  With the most halo suppression along with a type of unsharp masking near 0 - and gradually changes to mostly a type of deconvolution as the slider approaches 100.  So 50 would be half and half.
« Last Edit: February 05, 2017, 10:55:33 pm by Hoggy »
Logged
Cams: Pentax K-3, K-30 & Canon G7X, S100
Firm supporter of DNG, throwing away originals.
It's the hash, man..  That good hash!
Pages: [1]   Go Up