Ad
Ad
Ad
Pages: « 1 [2]   Bottom of Page
Print
Author Topic: Focus Magic returns...  (Read 4467 times)
Fine_Art
Sr. Member
****
Offline Offline

Posts: 1087


« Reply #20 on: November 19, 2013, 12:55:10 PM »
ReplyReply

What about fractal based resize? Onone's software perfect resize, which used to be Genuine Fractals, used to be the top resizing software. How is deconvolution impacted by fractal based interpolated pixels?

Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3672


« Reply #21 on: November 19, 2013, 01:21:31 PM »
ReplyReply

What about fractal based resize? Onone's software perfect resize, which used to be Genuine Fractals, used to be the top resizing software. How is deconvolution impacted by fractal based interpolated pixels?

Hi Arthur,

PR is supposed to create new detail, but that would be counterproductive for sharpening existing detail. When new detail is created on an up-sampled image, then that will potentially cause aliasing when the image is subsequently down-sampled.

The trick in up-sampling + deconvolving + down-sampling, is that initial blur >1 pixel (>3px after up-sampling), will be deconvolved/restored to approx. 1 pixel (3px before down-sampling), with relatively accurate pixel to pixel transitions.

When no detail is created in the upsampled image smaller than 3 pixels, then aliasing is almost impossible (with a proper down-sampling filter).

Cheers,
Bart
« Last Edit: November 19, 2013, 01:26:18 PM by BartvanderWolf » Logged
Fine_Art
Sr. Member
****
Offline Offline

Posts: 1087


« Reply #22 on: November 19, 2013, 03:24:21 PM »
ReplyReply

Good info, thanks once again.

Is upsizing, sharpen, downsizing superior to just doing the deconvolution in the raw processor?

I have 3 versions attached. This is from a 100% crop of a snow bank evenly littered with small leaves. I picked a crop spot with sharp focus at the bottom moving to de-focused at the top.

All are converted in Raw Therapee to 300% output then downsampled using bicubic in images plus.

One is unsharpened then deconvolved mildly ( I tend to prefer a natural, not a harsh look) in IP using adaptive Richardson-Lucy
One is sharpened in RT with their deconvolution sharpening settings radius .65,amount 65, damping 20, iterations 30
One is sharpened in RT with their deconvolution sharpening in addition their contrast by detail (which is also Richardson-Lucy) is also used press neutral, then contrast+, contrast+ giving a weighted to fine detail curve.

The CBD version starts to look crunchy at the bottom. The one with IP deconvolution is free of aliasing, downside is it has a bit of a soft focus look. Wedding photographers would probably go for this method. You can always run more iterations.

Its hard to say which is better all around. What do others think?
Logged
Hening Bettermann
Sr. Member
****
Offline Offline

Posts: 566


WWW
« Reply #23 on: November 19, 2013, 06:00:43 PM »
ReplyReply

Hi Bart,
with your and others' permission, a little flash back to posts # 16 and 17.
Do I understand your answer correctly to the end that when using Focus Magic, you don't use use your radius tool? Because as far as I can seen one can not use both, since FM does not allow entering the radius by the numbers.
Kind regards - Hening.
Logged

Hening Bettermann
Sr. Member
****
Offline Offline

Posts: 566


WWW
« Reply #24 on: November 19, 2013, 06:17:30 PM »
ReplyReply

to Fine_Art, post #22
I find #2 and 3 indistinguishable, and both seem to have better sharpness/micro contrast than #1.
Good light! - Hening.
Logged

Ligament
Full Member
***
Offline Offline

Posts: 178


« Reply #25 on: November 20, 2013, 01:15:00 AM »
ReplyReply

Hi,

You're welcome.

Unfortunately there are no plugins (that I know of) that do very good up-sampling and down-sampling, there's always something lacking. There are some that do good up-sampling for that specific purpose, e.g. Photozoom Pro or Blow-up, but they may create/add detail that will get in the way of deconvolution. For down-sampling most applications do not use linear gamma space to do the downsampling in, which can cause other color-blending issues, and many do not use proper down-sampling filters (e.g. Lanczos windowed Sinc or similar). Sigh.

What we really need, both for up-sampling and down-sampling, are better algorithms than most image editing applications have to offer. For superior quality we are forced to do it outside of Photoshop, and use e.g. ImageMagick, but that may be too disruptive in most workflows. ImageMagick allows to use the superior EWA resampling, which considers Elliptical/Circular regions for weighted interpolation, which gives very smooth up-sampled results that keep a lot of original detail intact, a very good starting point for deconvolution. EWA up-sampling works best in gamma adjusted space, so no need to first do a linearization for that.

With that in mind we can try to do slightly less optimal but still decent up-sampling with a Plug-in, but we need to anticipate deconvolution as the next step. One could use an automation Plugin like Photozoom Pro, with the S-Spline Max resize method, but with all three Fine-tuning controls zeroed out. Photozoom Pro is actually supposed to do up-sampling towards a final output end product, e.g. large format output, and for that it's very very good, because it actually adds edge resolution. But for an up-sample/down-sample round-trip we must dumb it down considerably, because the added resolution can cause aliasing artifacts when we down-sample, it adds a bit too much detail for a trouble free subsequent down-sampling. One just needs to try it on one's own images and subject matter.

It's a pity that we have to jump through so many hoops and use various plugins to do such a common task as restoring sharpness to our image captures, but for smaller volumes it is doable and the output quality can be improved a lot.

Cheers,
Bart


Dear Bart,

I attempted to install and use ImageMagick today on Mac OS 10.9, and found that the install process and usage was far beyond my UNIX skills, but thanks for mentioning it anyway.

I'll play around with PhotoZoom Pro, thanks.

Regarding down-sampling; which of the bicubic methods in photoshop do you recommend for the process we are discussing? Bicubic, Bicubic Smoother, Bicubic Sharper, Bicubic Automatic?

Many thanks again!
Logged
samueljohnchia
Sr. Member
****
Offline Offline

Posts: 293


« Reply #26 on: November 20, 2013, 07:38:25 AM »
ReplyReply

One is unsharpened then deconvolved mildly ( I tend to prefer a natural, not a harsh look) in IP using adaptive Richardson-Lucy
One is sharpened in RT with their deconvolution sharpening settings radius .65,amount 65, damping 20, iterations 30
One is sharpened in RT with their deconvolution sharpening in addition their contrast by detail (which is also Richardson-Lucy) is also used press neutral, then contrast+, contrast+ giving a weighted to fine detail curve.

Fine Art,

Interesting comparisons!

I find that No. 1 is the least crunchy looking, and No. 2 and 3 have visible halos, which I think should be considered an undesirable result in such relentless pursuit of sharpening perfection. 2 is a little stronger than 3. What first gave it away for me was the light "contour" of the edge bring slightly brighter. I then noticed that the dark was similarly darker, therefore stronger overall, but very close. Now that I think about it 2 and 3 are definitely unacceptable because of the very strong dark and light halos that feels unnatural. No. 1 is almost perfect except for maybe a tad too bright light halo, most evident in the edge of the darkish leaf in the lower right.

Thank you so much Bart, for illuminating the way once again with the most innovative methods to digital image processing I know of, from the average guy's perspective. You bridge the gap between the pioneering science and offer the layman a chance to grasp its sophistication with a great deal less effort. It is hugely appreciated!
« Last Edit: November 20, 2013, 07:41:26 AM by samueljohnchia » Logged
Fine_Art
Sr. Member
****
Offline Offline

Posts: 1087


« Reply #27 on: November 20, 2013, 11:17:37 AM »
ReplyReply

IP has a reduce artifacts button which greatly limits the halos. There is also another feature, you can put limits on the the luminance values in 16 bit that get deconvolved. I almost never use it, but you can wipe out white artifacts near 65535 or black ones near 0.
Logged
Fine_Art
Sr. Member
****
Offline Offline

Posts: 1087


« Reply #28 on: November 20, 2013, 11:55:39 AM »
ReplyReply

Here is 100 cycles of ARL, then adaptive contrast, then adaptive USM.

The ARL was limited in this run to not blow out the refection of the sun on the leaf bottom right. The range was 200-65355.
Downsampled bicubic.
Logged
Hening Bettermann
Sr. Member
****
Offline Offline

Posts: 566


WWW
« Reply #29 on: November 20, 2013, 04:30:04 PM »
ReplyReply

Quote from Samueljohnchia, post #26
> Thank you so much Bart, for illuminating the way once again with the most innovative methods to digital image processing I know of, from the average guy's perspective. You bridge the gap between the pioneering science and offer the layman a chance to grasp its sophistication with a great deal less effort. It is hugely appreciated!

+1!
« Last Edit: November 21, 2013, 05:01:09 PM by Hening Bettermann » Logged

BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3672


« Reply #30 on: November 21, 2013, 03:58:37 AM »
ReplyReply

Is upsizing, sharpen, downsizing superior to just doing the deconvolution in the raw processor?

It's hard to say in general, it depends on the source file and subject matter. However, when you run into problems with direct deconvolution, you may be able to still do something with the round-trip approach. Files from cameras without AA-filter may be problematic to sharpen, due to the aliasing that's present in the plane of optimal focus.

Quote
I have 3 versions attached. This is from a 100% crop of a snow bank evenly littered with small leaves. I picked a crop spot with sharp focus at the bottom moving to de-focused at the top.

All are converted in Raw Therapee to 300% output then downsampled using bicubic in images plus.

I assume that is with sharpening at the up-sampled stage?

Quote
Its hard to say which is better all around. What do others think?

On the sharpened image I see a reduction of some low contrast micro-detail and something that looks like a dark halo. Don't know what is causing that, but perhaps a somewhat lower smoothing, and different radius (seems a bit small, but then I have not measured your actual system PSF) will change that. An attempt with FocusMagic on your unsharpened (and mildly deconvolved) version gives quite a different result, retaining low contrast micro-detail while increased resolution.

FocusMagic usually gives very good results at 100% size already, but one can combine it with a Luminosity Blend-if layer which tends to make an even more subtle effect, tends to reduce noise, and avoids clipping issues, with the added possibility to use masks.

My alternative solution offers a possibility to apply deconvolution sharpening on up-sampled data, preferably without previous sharpening, so it can be used as capture sharpening at a sub-pixel level. The resampling algorithms should preferably be very good with minimal artifact creation. The benefit of using this method in combination with FocusMagic is that we because we cannot enter know blur radius information (or if we do not know it), we can use visual feedback in the small preview to hone in on the sub-pixel optimal level that matches the FocusMagic algorithms used.

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

Posts: 3672


« Reply #31 on: November 21, 2013, 04:04:51 AM »
ReplyReply

Hi Bart,
with your and others' permission, a little flash back to posts # 16 and 17.
Do I understand your answer correctly to the end that when using Focus Magic, you don't use use your radius tool? Because as far as I can seen one can not use both, since FM does not allow entering the radius by the numbers.

Hi Hening,

Because FocusMagic doesn't allow to enter values with decimal precision, it often requires to make a choice between 2 integer values for the optimal result, one slightly over the top, and one slightly under. However, with it's specific algorithms, it does a good job with such a coarse interval. Sometimes we do not know the optimal radius, because no test info is available, which is where an approach based on visual feedback can help, especially at an enlarged (sub-pixel) scale.

Cheers,
Bart
« Last Edit: September 09, 2014, 01:09:05 AM by BartvanderWolf » Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3672


« Reply #32 on: November 21, 2013, 05:37:54 AM »
ReplyReply

Regarding down-sampling; which of the bicubic methods in photoshop do you recommend for the process we are discussing? Bicubic, Bicubic Smoother, Bicubic Sharper, Bicubic Automatic?

Hi,

When we stay in Photoshop, Bicubic Smoother for up-sampling, and regular Bicubic for down-sampling will generally give the best compromise.

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

Posts: 3672


« Reply #33 on: November 21, 2013, 05:47:57 AM »
ReplyReply

Thank you so much Bart, for illuminating the way once again with the most innovative methods to digital image processing I know of, from the average guy's perspective. You bridge the gap between the pioneering science and offer the layman a chance to grasp its sophistication with a great deal less effort. It is hugely appreciated!

Hi Samuel,

You're welcome. I know that for many users the pixel peeking goes well beyond what they are willing to devote time on, but also that there are those who benefit from trying to explore the boundaries of what is possible. I usually try to seek (combinations of) methods that are easily accessible (and thus verifiable) to all, although I know that better results could be had if the industry just kept current with the developments and offered us convenient access from within the applications the sell for profit (nothing wrong with someone making profit, if everybody benefits). It's sometimes frustrating to see the slow pace.

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

Posts: 1087


« Reply #34 on: November 21, 2013, 10:37:01 AM »
ReplyReply

It's hard to say in general, it depends on the source file and subject matter. However, when you run into problems with direct deconvolution, you may be able to still do something with the round-trip approach. Files from cameras without AA-filter may be problematic to sharpen, due to the aliasing that's present in the plane of optimal focus.

I assume that is with sharpening at the up-sampled stage?

On the sharpened image I see a reduction of some low contrast micro-detail and something that looks like a dark halo. Don't know what is causing that, but perhaps a somewhat lower smoothing, and different radius (seems a bit small, but then I have not measured your actual system PSF) will change that. An attempt with FocusMagic on your unsharpened (and mildly deconvolved) version gives quite a different result, retaining low contrast micro-detail while increased resolution.

FocusMagic usually gives very good results at 100% size already, but one can combine it with a Luminosity Blend-if layer which tends to make an even more subtle effect, tends to reduce noise, and avoids clipping issues, with the added possibility to use masks.

My alternative solution offers a possibility to apply deconvolution sharpening on up-sampled data, preferably without previous sharpening, so it can be used as capture sharpening at a sub-pixel level. The resampling algorithms should preferably be very good with minimal artifact creation. The benefit of using this method in combination with FocusMagic is that we because we cannot enter know blur radius information (or if we do not know it), we can use visual feedback in the small preview to hone in on the sub-pixel optimal level that matches the FocusMagic algorithms used.

Cheers,
Bart

My guess would be that raw therapee does the upsizing (I picked lanczos) as a final step to the output. All sharpening in the program would then be capture sharpening. Maybe the devs on the forum can give the official answer.

I have been tempted to try focus magic several times. The only thing stopping me is I already have so many sharpening options.
Logged
Ligament
Full Member
***
Offline Offline

Posts: 178


« Reply #35 on: November 21, 2013, 11:07:53 AM »
ReplyReply

Hi,

When we stay in Photoshop, Bicubic Smoother for up-sampling, and regular Bicubic for down-sampling will generally give the best compromise.

Cheers,
Bart

Thank you so much for answering all our questions and sharing your knowledge Bart!
Logged
BartvanderWolf
Sr. Member
****
Offline Offline

Posts: 3672


« Reply #36 on: November 21, 2013, 12:51:08 PM »
ReplyReply

Thank you so much for answering all our questions and sharing your knowledge Bart!

Hi,

I'm glad to be of help.

Cheers,
Bart
Logged
Pages: « 1 [2]   Top of Page
Print
Jump to:  

Ad
Ad
Ad