Ad
Ad
Ad
Pages: « 1 ... 3 4 [5]   Bottom of Page
Print
Author Topic: sharpening-Lightroom vs PhotoKit in Photoshop or both?  (Read 22688 times)
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3580


« Reply #80 on: February 18, 2010, 04:30:45 AM »
ReplyReply

Quote from: Schewe
Unless you've done extensive tests using either Lightroom 2.3 or Camera Raw 5.3 then you are not judging what I consider to be optimal resampling of raw files (although again, I don't like the crude size and resolution controls in Camera Raw).

Hi Jeff,

So then, how does a downsample from 5616 to 800 pixels high look with and without prior (capture) sharpening?
I've supplied 2 crops, both with and without sharpening, so that should be easy to test for those with Lightroom 2.3 or Camera Raw 5.3.

I encourage everybody to conduct the test themselves (Crop1 is probably most challenging) and compare the result with the Photoshop bicubic sharper samples I've already supplied earlier. As suggested, in Photoshop just increase the canvas of the downloaded crop to 5616 pixels high and 3744 pixels wide, or alternatively use the full size sharpened JPEG, and import that into ACR 5.3 as a raw or use LR 2.3 and try to downsample. I'm looking forward to the results and procedure used.

Cheers,
Bart
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3580


« Reply #81 on: February 18, 2010, 04:54:51 AM »
ReplyReply

Quote from: jbrembat
Bart, you are on the wrong way.
The problem, on downsizing, is not the interpolation quality:
to avoid aliasing image have to be filtered before size reduction.

Hi Jacopo,

I disagree, the problem is very much the interpolation quality, which should use proper filters (or even offer a choice for different image content).

Here I'm specifically challenging the conventional mantra that one always needs to (capture) sharpen at the original size before output sharpening at the final size. Without the proper precautions such as the use of a good resampling algorithm (which automatically pre-filters the proper high spatial frequencies), pre-sharpening does not help. Of course pre-blurring would help, but first sharpening and then needing to preblur doesn't look like an efficient workflow and even more troublesome in finding the correct blur amount. Besides, the question remains, how to blur without losing too much resolution in the end result? A good filter is not available in the toolset, so one needs a better method/algorithm. Does Lightroom produce better results than Photoshop?

Another challenge is how to capture sharpen without creating something that hurts the interpolated/enlarged output. Regular sharpening techniques require a lot of attention to avoid halo creation and don't improve resolution but increase edge contrast, deconvolution restoration does much better. We don't want to enlarge artifacts, do we?

Cheers,
Bart
« Last Edit: February 18, 2010, 04:56:53 AM by BartvanderWolf » Logged
jbrembat
Full Member
***
Offline Offline

Posts: 177


« Reply #82 on: February 18, 2010, 07:36:04 AM »
ReplyReply

Quote from: BartvanderWolf
Hi Jacopo,

I disagree, the problem is very much the interpolation quality, which should use proper filters (or even offer a choice for different image content).

Here I'm specifically challenging the conventional mantra that one always needs to (capture) sharpen at the original size before output sharpening at the final size. Without the proper precautions such as the use of a good resampling algorithm (which automatically pre-filters the proper high spatial frequencies), pre-sharpening does not help. Of course pre-blurring would help, but first sharpening and then needing to preblur doesn't look like an efficient workflow and even more troublesome in finding the correct blur amount. Besides, the question remains, how to blur without losing too much resolution in the end result? A good filter is not available in the toolset, so one needs a better method/algorithm. Does Lightroom produce better results than Photoshop?

Another challenge is how to capture sharpen without creating something that hurts the interpolated/enlarged output. Regular sharpening techniques require a lot of attention to avoid halo creation and don't improve resolution but increase edge contrast, deconvolution restoration does much better. We don't want to enlarge artifacts, do we?

Cheers,
Bart
Quote
such as the use of a good resampling algorithm (which automatically pre-filters the proper high spatial frequencies)
Yes, a pre-filter must be applied. So you don't need a better resampling algorithm, you need a prefilter before resampling.
Of course this may be performed when you click on a button.
But the only way to do the job is:
1- filter the image
2- resample the filtered image (this is the resampling algorithm)

For example in this thread was said that using Lanczos interpolation you don't get aliasing, that's not true at all.
Lanczos interpolation is a windowed sinc interpolation and if no pre-filter is performed, it genarates aliasing.
Any resampling algorithm generates aliasing. Some algorithms give a smoother result that masks more artifacts, but it is not enough for strong image size reductions.

Jacopo

Logged
joofa
Sr. Member
****
Offline Offline

Posts: 486



« Reply #83 on: February 18, 2010, 09:46:49 AM »
ReplyReply

Quote from: jbrembat
But the only way to do the job is:
1- filter the image
2- resample the filtered image (this is the resampling algorithm)

For example in this thread was said that using Lanczos interpolation you don't get aliasing, that's not true at all.
Lanczos interpolation is a windowed sinc interpolation and if no pre-filter is performed, it genarates aliasing.
Any resampling algorithm generates aliasing. Some algorithms give a smoother result that masks more artifacts, but it is not enough for strong image size reductions.

The prefiltering may not be seen as a separate process in downsampling. For integer downsampling ratios it automatically comes into play when the expanded downsampling kernel is used. However, for interpolation schemes when used for downsampling, such as BiCubic, it messes up producing aliasing. However, a way to avoid that is what I pointed out in post # 63 on how to convert a particular BiCubic interpolation procedure, derived from Keys interpolation, into a coupled downsampling kernel.
« Last Edit: February 18, 2010, 10:46:28 AM by joofa » Logged

Joofa
http://www.djjoofa.com
Download Photoshop and After Effects plugins
jbrembat
Full Member
***
Offline Offline

Posts: 177


« Reply #84 on: February 18, 2010, 11:06:15 AM »
ReplyReply

Quote from: joofa
The prefiltering may not be seen as a separate process in downsampling. For integer downsampling ratios it automatically comes into play when the expanded downsampling kernel is used. However, for interpolation schemes when used for downsampling, such as BiCubic, it messes up producing aliasing. However, a way to avoid that is what I pointed out in post # 63 on how to convert a particular BiCubic interpolation procedure, derived from Keys interpolation, into a coupled downsampling kernel.

In any case you have to filter before interpolating.
For performance reason, the filter and the interpolation may be executed first on x  and then on y. So filter have to be applied before.

Jacopo
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3580


« Reply #85 on: February 18, 2010, 11:14:34 AM »
ReplyReply

Quote from: jbrembat
Yes, a pre-filter must be applied. So you don't need a better resampling algorithm, you need a prefilter before resampling.
Of course this may be performed when you click on a button.
But the only way to do the job is:
1- filter the image
2- resample the filtered image (this is the resampling algorithm)

As Joofa also mentioned, "The prefiltering may not be seen as a separate process in downsampling" , a coupled kernel can be used.

Both 1. and 2. are needed, we agree on that, I consider them integral to a proper downsampling. Prefiltering without a proper downsampling algorithm will still result in poor quality results. So I maintain that a good resampling algorithm must be used.

Cheers,
Bart
« Last Edit: February 18, 2010, 11:18:53 AM by BartvanderWolf » Logged
joofa
Sr. Member
****
Offline Offline

Posts: 486



« Reply #86 on: February 18, 2010, 11:18:31 AM »
ReplyReply

Quote from: jbrembat
In any case you have to filter before interpolating.
For performance reason, the filter and the interpolation may be executed first on x  and then on y. So filter have to be applied before.

Here is the issue, and we shall keep the discussion simple for integer downsampling ratios. As the downsampling ratio increases the extent of any proper downsampling kernel increases to more and more pixels so the prefiltering is already built into this process, and you don't need any extra prefilter. On the other hand if one uses a usual interpolation-based mechanism for downsampling, then as the downsampling ratio increases your window to work with the pixels may be held constant (perhaps this is how Photoshop does it in BiCubic as Schewe has hinted above) and that is why aliasing happens, and to avoid that you use a prefilter, in such interpolation-downsampling schemes. People use heuristics in such methods on which/how prefilter to use in this situation. In post # 63 I mentioned that one may not have to use such heuristics and may derive an appropriate prefilter, coupled with some interpolation-downsampling mechanism, using some optimization criterion.

Logged

Joofa
http://www.djjoofa.com
Download Photoshop and After Effects plugins
bjanes
Sr. Member
****
Online Online

Posts: 2783



« Reply #87 on: February 18, 2010, 12:13:33 PM »
ReplyReply

Quote from: BartvanderWolf
I use a couple of different ones, but FocusMagic works fine on a 32bit hardware platform. Unfortunately it is not ready for 64-bit hardware/software, and I'm not sure whether they will update. A friend of mine does have FocusMagic running on his 64-bit Windows 7 PC, but only when running the 32-bit version of Photoshop CS4. On my Vista Ultimate 64-bit system it won't install. I also use ImagesPlus 64-bit version which allows to specify the PSF (upto a 9x9 kernel) for use with e.g. the adaptive Richardson Lucy restoration or a few others. A free version of the RL algorithm can be found as an alternative sharpening tool in the RawTherapee converter, but it is not possible to define one's own PSF so I assume RT uses a Gaussian as PSF (through its radius control). RL is also implemented in MatLab.
Bart,

Thanks for the update. FocusMagic installed normally on my 64 bit Win 7 machine and works fine with the 32 bit version of Photoshop. I understand that MatLab is relatively expensive and is used by scientists and has a steep learning curve. R-L is also available in Iris, but the PSF is derived by selecting a point source (a star). With a terrestrial image lacking such a source, one could try to insert one from another image, but I haven't tried this. For less sophisticated users who can't or don't want to experiment at length with PSFs, FocusMagic looks like a good deconvolution method.

FWIW, I did a few experiments with ACR sharpening compared to FocusMagic using Imatest and images from the Nikon D3 with the 60 mm AFS f/2.8 micro. I merely used autofocus, so the focus may not be optimum but the image can be used for comparison. To keep things simple I used the ACR sharpening defaults (amount = 25, radius = 1.0 and detail = 25). For FM I experimented interactively and used a radius of 2 and amount 100. The results are shown.

Others can comment, but the essential points that I would make are as follows. One measure of aliasing is the MTF at Nyquist. It is lowest with no sharpening, but is increased with ACR and FM sharpening. The MTF in the midfrequencies is low with no sharpening, but is improved with ACR and FM sharpening, more so with the latter. The effect can be seen visually on the shown actual images.  One could try some more optimal ACR sharpening settings, but the default looked good on the preview and further experimentation is left to a later time, perhaps when the improved demosaicing and sharpening routines become available in the next version of ACR (similar to what is present in the LR beta).

Bill

[attachment=20339:003_ACR_..._YA2_cpp.png][attachment=20337:003_ACR_...BR12_cpp.
png] [attachment=20341:003_ACR_...BR12_cpp.png][attachment=20342:CompositeSh.png]
Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6899


WWW
« Reply #88 on: February 18, 2010, 12:21:15 PM »
ReplyReply

Bill,

If you have the time and interest, I'd really be interested in seeing the results of these alternatives on real-world high-frequency photographs.

Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3580


« Reply #89 on: February 18, 2010, 01:20:25 PM »
ReplyReply

Quote from: Mark D Segal
Bill,

If you have the time and interest, I'd really be interested in seeing the results of these alternatives on real-world high-frequency photographs.

Mark,

May I suggest downloading my example crops. You'll have an unsharpened and a FocusMagic restored result, you can duplicate the unsharpened layer and apply your preferred sharpening method and compare the layers quite easily.

Cheers,
Bart
Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6899


WWW
« Reply #90 on: February 18, 2010, 01:36:58 PM »
ReplyReply

Quote from: BartvanderWolf
Mark,

May I suggest downloading my example crops. You'll have an unsharpened and a FocusMagic restored result, you can duplicate the unsharpened layer and apply your preferred sharpening method and compare the layers quite easily.

Cheers,
Bart

Been there, done that.....see post #77

Cheers,

Mark
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3580


« Reply #91 on: February 18, 2010, 02:09:02 PM »
ReplyReply

Quote from: bjanes
Bart,

Thanks for the update. FocusMagic installed normally on my 64 bit Win 7 machine and works fine with the 32 bit version of Photoshop.

Hi Bill,

Thanks for the feedback. Guess I'm going to upgrade to Win 7 as well then ...

Quote
FWIW, I did a few experiments with ACR sharpening compared to FocusMagic using Imatest and images from the Nikon D3 with the 60 mm AFS f/2.8 micro. I merely used autofocus, so the focus may not be optimum but the image can be used for comparison. To keep things simple I used the ACR sharpening defaults (amount = 25, radius = 1.0 and detail = 25). For FM I experimented interactively and used a radius of 2 and amount 100. The results are shown.

Thanks, you just beat me to it ;-)  The results look as expected. In hindsight perhaps a slightly lower 'amount' setting for FM would have reduced the slight overshoot/halo, but this can also be addressed with a Blend-if setting of the sharpening layer. The FM radius=2 setting looks spot on, it produces as steep an edge transition as one could reasonably expect. As reference I here add how a 'perfect' edge would look:
[attachment=20346:PerfectE...pc_Y_cpp.png] [attachment=20355:PerfectE...pc_Y_cpp.png]

It's from a synthetical ideal slanted edge without any resizing or sharpening, the first one interpreted as a linear gamma image, and the second one as a gamma 0.5 (inverse gamma 2.0) image.

Cheers,
Bart
« Last Edit: February 18, 2010, 02:47:45 PM by BartvanderWolf » Logged
Eric Brody
Full Member
***
Offline Offline

Posts: 106


WWW
« Reply #92 on: February 19, 2010, 09:44:33 PM »
ReplyReply

WOW!!! I asked what I thought was a reasonably simple question... and all hell broke loose with over 90 replies, most of which went completely over my head! And I went to 23rd grade! What an incredible forum.

I get a sense that the sharpening routines in Lightroom are the basic equivalent of Photokit except for the "creative" sharpening aspect and that output sharpening part can be done in Lightroom, or with Photokit. Is that the jist of these 90 replies?

Frankly neither do I care about nor do I understand  "deconvolution," and many of the other complexities discussed in this thread.

Most sincere thanks to everyone for all their comments. Now all I have to do is finally decide what to do. Can I learn how to sharpen in Lightroom? I hope so. It's a lot cheaper than the hundred dollars for Photokit though I have the trial and will see if I want to actually purchase it.

Eric
Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6899


WWW
« Reply #93 on: February 19, 2010, 10:43:36 PM »
ReplyReply

Quote from: brodyer
WOW!!! I asked what I thought was a reasonably simple question... and all hell broke loose with over 90 replies, most of which went completely over my head! And I went to 23rd grade! What an incredible forum.

I get a sense that the sharpening routines in Lightroom are the basic equivalent of Photokit except for the "creative" sharpening aspect and that output sharpening part can be done in Lightroom, or with Photokit. Is that the jist of these 90 replies?

Frankly neither do I care about nor do I understand  "deconvolution," and many of the other complexities discussed in this thread.

Most sincere thanks to everyone for all their comments. Now all I have to do is finally decide what to do. Can I learn how to sharpen in Lightroom? I hope so. It's a lot cheaper than the hundred dollars for Photokit though I have the trial and will see if I want to actually purchase it.

Eric

Eric, I sympathize with you completely. YES, sharpen in Lightroom - it's fine! And it's not rocket-science. Here's a practical guide for how to apply it:

The image needs to be at 100% display magnification to see the effect. You get a better impression using the B&W feature which shows in the little preview window above the sliders by doing ALT-Click on the individual sliders.

Radius: Lower values give more narrow edge detail. Use values below 1 for high frequency images and values above 1 for lower frequency images. (Frequency refers to the predominance of fine edge detail).

Detail: It distinguishes edges from halos. Lower values suppress halos and allow the use of a higher Amount for mor edge sharpening; higher values accentuate halos and show more of the sharpening effect.

Masking: Increasing it blocks out more area of lesser interest for sharpening. What is masked shows as black in the little preview window with ALT pressed when you slide the Masking control.

Amount: The extent to which detail and radius are applied. Higher amounts apply them more strongly.

That's it.

The only way to master this is to examine what you are doing on the display (experiment with several types of images), note the numbers you use, and print the image. Look at the results on paper. Experiment until you develop prototypical settings which give you the appearance in print you like.

More often than not I simply use the Presets in Lightroom for sharpening - Landscape and Portrait, because on the whole, when you apply them to those respective kinds of images they really do deliver nice results.

For output sharpening in the print module - it's quite mechanical. There are only two settings, one for paper type and one for strength - again do some experiments to see which combination looks best to you ON PAPER (note the PPI setting above the sharpening should be the PPI at which you want to print). The display is not a reliable place to judge the final impact of sharpening.
« Last Edit: February 19, 2010, 10:47:47 PM by Mark D Segal » Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
ErikKaffehr
Sr. Member
****
Online Online

Posts: 7333


WWW
« Reply #94 on: February 20, 2010, 02:31:30 AM »
ReplyReply

Hi,

The answer is a bit complex. There is little difference in capture sharpening and output sharpening in PKS and Lightroom. PKS is using PS-tools while Lightroom implements the same ideas using specially developed algorithms, which probably are better and definitively faster.

Also, it may be better to have a Lightroom based parametric workflow. Working in PS breaks parametric workflow and increases file sizes. Don't use PS if you don't need it.

The part where PKS really shines, in my book at least, is creative sharpening. There are nice methods for simulating depth of field, cutting haze and so on. PKS may also be nice if you work with images scanned on film.

Best regards
Erik




Quote from: brodyer
WOW!!! I asked what I thought was a reasonably simple question... and all hell broke loose with over 90 replies, most of which went completely over my head! And I went to 23rd grade! What an incredible forum.

I get a sense that the sharpening routines in Lightroom are the basic equivalent of Photokit except for the "creative" sharpening aspect and that output sharpening part can be done in Lightroom, or with Photokit. Is that the jist of these 90 replies?

Frankly neither do I care about nor do I understand  "deconvolution," and many of the other complexities discussed in this thread.

Most sincere thanks to everyone for all their comments. Now all I have to do is finally decide what to do. Can I learn how to sharpen in Lightroom? I hope so. It's a lot cheaper than the hundred dollars for Photokit though I have the trial and will see if I want to actually purchase it.

Eric
« Last Edit: February 20, 2010, 02:32:11 AM by ErikKaffehr » Logged

bjanes
Sr. Member
****
Online Online

Posts: 2783



« Reply #95 on: February 20, 2010, 07:30:01 AM »
ReplyReply

Quote from: Mark D Segal
Bill,

If you have the time and interest, I'd really be interested in seeing the results of these alternatives on real-world high-frequency photographs.
Mark,

Here is a landscape shot with the D3 and 60 mm f/2.8 AFS at f/8 and viewed at 100%. It is not as sharp as Bart's windmill image, but I did use a tripod. It was shot at a relatively close distance and the subject distances, front to back, vary. I can't remember where I set the focus point with auto focus and I didn't use live view or mirror lockup. Nonetheless, the image is typical for what might be obtained with this camera under field conditions. I capture sharpened with ACR using Amount = 100, radius = 0.6, and Detail = 25 and with FocusMagic Radius = 2, amount = 100. I can see little difference in the images. The FM sharpening is on the left and the ACR on the right.

Regards,

Bill

[attachment=20383:CBG__081...7_fm_acr.jpg]
Logged
Mark D Segal
Contributor
Sr. Member
*
Offline Offline

Posts: 6899


WWW
« Reply #96 on: February 20, 2010, 08:32:54 AM »
ReplyReply

Quote from: bjanes
Mark,

Here is a landscape shot with the D3 and 60 mm f/2.8 AFS at f/8 and viewed at 100%. It is not as sharp as Bart's windmill image, but I did use a tripod. It was shot at a relatively close distance and the subject distances, front to back, vary. I can't remember where I set the focus point with auto focus and I didn't use live view or mirror lockup. Nonetheless, the image is typical for what might be obtained with this camera under field conditions. I capture sharpened with ACR using Amount = 100, radius = 0.6, and Detail = 25 and with FocusMagic Radius = 2, amount = 100. I can see little difference in the images. The FM sharpening is on the left and the ACR on the right.

Regards,

Bill

Bill, thanks for doing that, and yup - I agree - based on what's showing on my display I'd say they're within a pixel-peep of each other.

Cheers,

Mark
Logged

Mark D Segal (formerly MarkDS)
Author: "Scanning Workflows with SilverFast 8....." http://www.luminous-landscape.com/reviews/film/scanning_workflows_with_silverfast_8.shtml
VictorBushkov
Newbie
*
Offline Offline

Posts: 1


« Reply #97 on: June 21, 2010, 12:13:41 AM »
ReplyReply

I won't start another topic, because I have a somewhat related question. If I want to output an image for web, I downsample it through LR export dialog (since as I understand from the discussion downsampling in LR works better than in PS). In the dialog I can choose output sharpening for a screen with three options (low, standart, high). So the question is the following. Does the algorithm for output sharpening in LR take into account image content like the algorithm in PKS where it's possible to choose different image content (SuperFine, Narrow and etc)? Or it is better to downsample an image in LR without output sharpening and then sharpen the resulting image in PS with Photo Kit output sharpening?

And does ACR 6.1 use the same algorithms as LR 3?
« Last Edit: June 21, 2010, 12:18:02 AM by VictorBushkov » Logged
Pages: « 1 ... 3 4 [5]   Top of Page
Print
Jump to:  

Ad
Ad
Ad