Where is the original? I must be looking at the wrong messages.
-
Welcome to the new forums! Please read this first. For known issues we are working to resolve, click here.
Optical Spatial Frequency Filtering of Image Sensors ?
- Thread starter Detail Man
- Start date
-
- Tags
- nikon d800e
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
DxO Labs has always ben very tight-lipped about their secret-sauce. However, when I read the statements made here:
http://www.dxo.com/us/photo/dxo_optics_pro/features/optics_geometry_corrections/lens_softness
... it appears that the deconvolution-deblurring is integrated into the processes - and perhaps so on a level where the deconvolution is different for each of the RGB channels (on a RAW, pre-demosaiced level, I suspect, but do not know for sure). I do not know if any separate "edge-enhancement" (as in USM-related) processes are employed. I would hope not. That would seem to defeat the intention - as it seems (to me) Adobe apps manages to in their own (parallel-path summation) manner.
Well, Jack, there are a number of ways that I think that your intended comparison could be better accomplished, and some basic points of concern:Here is a first pass: the original, DxO with your settings and InFocus alternating in this sequence every two seconds. For my tastes the DxO settings are too strong (fuzzy tonal-contrast) so I matched them with the 'Pre-sharpening Common' setting in InFocus that I also usually consider too strong (too crisp). I don't know what DPR will do to the GIF so I recommend downloading the original and opening it in a browser - hit ctrl-zero to make sure that the browser is not zooming the images: in these comparisons one is not allowed to view images at more than 100% ;-)
Original at 100%, DxO as setup by DM, Topaz InFocus 'Presharpening Common' settings
Below are just the DxO and InFocus renderings alternating every 3 seconds for ease of comparison. It's clear which is which.
DxO and InFocus set up as above
Jack
Edit: It appears that DPR is not displaying the animated GIFs properly (the top one just looks like the original and the bottom one just looks like InFOcus). Let me know if the originals work (they work for me). If they don't I'll post individual images.
Do we know if InFocus proceeds on a RAW-data level in the Adobe apps that it runs in ? If that is indeed the case (which would make it potentially more effective as an application), then that alone is the representative way to proceeed - and TIF-level approaches are not representative of InFocus' (at least potential) performance ...
You (must) have proceeded (in the case of both applications) on a TIFF level only. I have offered to email you the 16-bit TIF from which my (100% crop) JPG format posted sample (in my DPR Gallery) was created. Any TIF that you could make from that JPG is 8-bit only (which seems less than ideal as a starting-point).
Further, while the JPG was created using no Chroma Sub-sampling, and a 100% Quality Factor (thus, no loss-less processes, and the Huffman encoding is lossless), there are still computational quantization errors associated with the process - which manifest in the fact that a JPEG decoding (as you must have performed) has an average 8-bit tone-level error of around 5 tone-levels out of 255 (2%). See the final statement in this section of this web-page:
http://en.wikipedia.org/wiki/JPEG#Decoding
Because your animated GIFs are a larger pixel-dimension than 940 (in either dimension), the DPR display-system has simply grabbed a single (seemingly arbitrary) frame and performed the standard server-resource-saving "meat job" on it. That would be applying 2x2 CSS and a 90% QF to a vastly smaller (in Bytes) re-encoded JPG display image.
Clicking on the "original" link, I can see your down-sampled versions is my browser. The first one alternates between "DxO Lens Softness" and no "DxO Lens Softness". I am absolutely sure that the second one displays 3 (not 2) different images. It appears to include the same two representations as the first image - and a third representation (that is InFocus?). I don't know. Please have a close look at what you created, and advise. I myself would rather see you post separate originals that I can place in a quality image-viewer of choice and compare manually.
We all have our own vernaculars with which to attempt to describe our perceptions. I cannot say that your phrase "fuzzy tonal-contrast" makes much direct sense to me. Can you explain more precisely what you mean by that ? It seems a rather strange juxtaposition of descriptors (to me).
What is InFocus' "Pre-sharpening Common", and what does it do (and where in the processing-chain do you think that it does it) ?
I'm not particularly concerned with which is the (in any particular viewer's mind) the "winner". It's implicitly subjective, anyway. But it would make sense that any comparison be as uniform and free from questions regarding methods as possible. For instance, unless InFocus proceeds (only) on a 16-bit TIFF level, such a comparison would as a result not do InFocus justice.
Rather than an "either/or" dichotomy ruling the day, it may be that both applications have their particular merits and weaknesses - and I am interested in them.
For instance, DxO Optics Pro (through Versions 7.x, anyway) have not included any meaningful non-linear "highlight recovery" functionality. This has been a bit of a mystery to me in the past - but some recent experiences with blown Green RAW channels (at the top of the trio of tall bushes in the far-field) proved to be problematic when applying "Lens Softness" corrections in this image (which also had some very, very mild post-USM applied after DxO in PSP X4, 16-bit):
Download Original at: http://www.dpreview.com/galleries/4464732135/download/2228225
I am now of a line of thinking that DxO Optics Pro very likely (as suspected) accomplishes the "Lens Softness" corrections (at least to some extent) at the RAW-data, pre-demosaiced, level - and that it may well be problems like this which post de-mosaicing non-linear tone-curve transfer-function trickery (like "highlight recover") is not able to effectively address ...
The foliage-detail in the 100% crops posted that we are considering have blown skies in RAW, but the foliage-detail is definitely not blown in any of the RAW RGB-channels, however.
This raises the question about how and where in the signal-processing chain InFocus operates. 16-bit TIFF level could be both a benefit and a weakness. Interested to (try to) know.
.
Would be happy to email you the source 16-bit TIF, so that you do not have to try to decode a JPG. Just PM me an email address. Once MIME-encoded, it should be around 21.3 Mbytes attachment-size. Most email services allow (at least) 25 Mbyte total message-size these days.
Could also send you the GH2 RW2 (which also just fits within most 25 mbyte limits). Obviously, you would then want to install the DxO 8.x trial, and process the RAW in each application as intended in order to be able to create what we know to be an even and representative comparison.
The whole idea of you or other readers drawing premature and essentially uniformed conclusions from anything (potentially) less than an even and representative comparison strikes me as perhaps causing more potential harm than good. As well, this image is indeed of the kind that we might like to use in order to evaluate a deconvolution-deblurring tool. Portraits seem way off the mark - unless one wants to pixel-peep at eyebrows and eyelashes, I suppose ...
DM ...
Attachments
Last edited:
If the green is clipped in raw, what do you expect any software to be able to do in this instance? Highlight recovery can work where there is still data available in raw, but if it's clipped in raw, the best it can do is to extrapolate from surrounding data. In the case of these bushes, the surrounding area is the near black deep shadow of the maple tree. Seems that the original was overexposed. You'd probably have been happier having shot with a polarizer. That would cut those near specular highlights yet leave the remainder of the image the same.
--
Victor Engel
Loading…
victorspictures.com
Also please check out my new game blog
Loading…
swarmsgame.blogspot.com
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
Sure. I don't expect that much of anything (necessarily) can be done - if the deconvolution deblurring occurs prior to RAW-level demosaicing, that is. Post-demosaicing (could) be a different story ...
My point was more as to where in the processing-chain DxO's deconvolution-deblurring may be being performed. I found that adding a tiny (but adequate in terms of reported RGB tone-levels) soft-clip at the top in it's (post-demosaiced) Tone Curve tool appeared to do very little if anything to improve the appearance when "Lens Softness" was activated. Ergo, the "Lens Softness" (may) be performed on the RAW pre-demosaiced level. That's all ...
That (may) be a reason why DxO Labs has (curiosly) not (in Versions through 7.x) include a "highlight recovery" widget to compete with Adobe's recent much lauded (but found in a MFT forum thread to appear to me to be inferior to RT's and RPP's) "highlight recovery" widget.
The DxO Optics Pro 8.x "marketing sizzle" appears to imply that they have now dived into that ("highlight recovery widget") business. Don't know. I am sticking with DxO Optics Pro Version 7.23 that I know works well on my WinXP Pro 32-bit OS, and works with my GH2 and LX3 RAWs.
DxO Labs are jerks about trying to force users to endlessly upgrade to ever-bloating and ever-changing updates/upgrades - by intentionally (and technically arbitrarily, I might add) wiring things so that newer released DxO Optical Corrections Modules will not function.
After purchasing a Version 7.x license to play their profit-gathering-game, they alienated me by making Versions 7.5+ a primarily 64-bit item (which they boldy claim is not a bloated pig on my 32-bit system, though I have little faith in such assurances coming out of them in particular), and requiring that I install it in order to be able to use newer DxO OCMs.
Perhaps fair game in the case of a brand new release. Rather a "stick in the eye" to pull on users who have purchased a version license (i.e., Versions 7.x) and expect functionality ...
I've little doubt that the same game applies to Versions 8.x (released less than 1 year after Version 7.0; likely as buggy and ragged for a while as Versions 6.x and 7.x were). I'll stay with Version 7.23.
Maybe so.
Last edited:
Jack Hogan
Veteran Member
InFocus does not have access to the raw data and works on the rendered image only. In this case it worked on your admittedly limiting jpeg: PM with my email address sent for 16 bit TIFF to be used instead.
They should be 100% crops of your jpeg, with no downsampling other than GIF color encoding. The sequence of the three image animation is 'original', 'DxO', InFocus. You are not allowed to look at them at more than 100% so hit ctrl-zero on your browser to make sure it is not zooming the images
But perhaps we should wait for the TIFF to perform a proper comparison.Topaz recommends using InFocus mostly for Capture Sharpening, and I agree. You can play with deconvolution and micro-sharpening dials yourself, but InFocus has three standard presets of increasing strength: pre sharpening minor, pre-sharpening common, pre-sharpening detailed. I almost never use them, but for repeatability (even by third parties) I used the middle one, 'common' - which as I mentioned is usually too strong of an effect for my taste - on your image.
My workflow for images where I would like to maximize capture sharpness is raw development in CNX2 (Nikon shooter) with a minimum of built-in sharpening (to take advantage of Nikon's pre-rendering routines) and then pass the 16 bit rendered image to InFocus' Auto function, which scans the image and determines deconvolution parameters automatically giving much more conservative results than those of the presets.
100% agreed. I am just curious to see what the state of the art is since the last time I looked at this topic. If possible, let's see if we can isolate the deconvolution component of DxO - it's easy to do in InFocus.
Jack
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
Got it, and one pair out of 3 pairs of 100% 16-bit TIF crops are in your email inbox. The first is from the center area of the frame (where things looks best). Will send a crop around the right-frame area that you have already tried, and a crop around the left-frame area (which looks to be little bit less soft, presumably due to the variation in the LGV 7-14mm lens-system itself).
Hmmm. Interesting. I wonder how it makes enough sense out of the image-data to somehow estimate the width of (presumably a Gaussian) PSF ? Or, do you think that it does something else ?They should be 100% crops of your jpeg, with no downsampling other than GIF color encoding. The sequence of the three image animation is 'original', 'DxO', InFocus. You are not allowed to look at them at more than 100% so hit ctrl-zero on your browser to make sure it is not zooming the images
Topaz recommends using InFocus mostly for Capture Sharpening, and I agree. You can play with deconvolution and micro-sharpening dials yourself, but InFocus has three standard presets of increasing strength: pre sharpening minor, pre-sharpening common, pre-sharpening detailed. I almost never use them, but for repeatability (even by third parties) I used the middle one, 'common' - which as I mentioned is usually too strong of an effect for my taste - on your image.
My workflow for images where I would like to maximize capture sharpness is raw development in CNX2 (Nikon shooter) with a minimum of built-in sharpening (to take advantage of Nikon's pre-rendering routines) and then pass the 16 bit rendered image to InFocus' Auto function, which scans the image and determines deconvolution parameters automatically giving much more conservative results than those of the presets.
Not quite sure what you mean by "isolate the deconvolution component of DxO" ? Can you explain ?
I see that Topaz makes some pretty ambitious and rosy claims about InFocus - claims that are not made surrounding deconvolution algorithms (including DxO Labs "Lens Softness"):
Topaz InFocus uses this same deconvolution technology to mathematically reverse image blur while increasing the actual sharpness. InFocus can enhance the clarity of an already well-focused as well as deblur an out-of-focus or motion blurred image. The ability to re-focus a blurred image (after the fact) is a challenging but valuable capability. This technology can be applied to both enhancing the clarity in an already well-focused image, as well as de-blurring out-of-focus or motion blurred images.
A common side effect of the deconvolution process is deconvolution artifacts. These artifacts can appear as a collection of artifacts around lines of details in an image. Topaz InFocus includes a parameter specially designed to eliminate these artifacts.
http://www.topazlabs.com/blog/topaz-infocus-deconvolution-technology/
Wonder what part of those claims are real ? What are your impressions from using it ? Do you think that the "pre-sharpening" uses some kind of edge-enhancement processes ? Sort of sounds like it may do something like that.
I ususally (but not always) apply some very mild post-USM (16-bit, using PSP X4) following 16-bit Lanczos-3 down-sampling of the 16-bit TIFs that DxO outputs. I do find doing that generally, usually preferable to "goosing" the "Lens Softness" settings. However, when I try to deal with a somewhat de-focused or perhaps motion-blurred shot, using higher "lens Softness" setting can work better than using higher amounts of post-USM.
All in all, though (as I understand it), deconvolution-deblurring has not generally been considered by most to be an effective tool for dealing with such de-focused and motion-blurred images.
Here is a fellow who is working on some algorithms to try to deal with de-focused subject-matter:
http://yuzhikov.com/articles/BlurredImagesRestoration1.htm
Then there is Jue Wang working for Adobe on similar pursuits (crowed about in Oct 2011):
http://juew.org/deblurFamousPhoto.html
Last edited:
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
As I usually apply mild 16-bit USM following RAW processing with DxO Optics Pro using "Lens Softness" corrections (in addition to DxO's other optical corrections applied), here are "without Lens Softness corrections or post-USM" followed by "with Lens Softness corrections and post-USM" loss-less JPGs made from the 100% TIF crops that I have emailed to Jack Hogan (who will be presenting the "Lens Softness" processed versions as loss-less JPGs without any post-USM applied).
Because the maximum pixel dimension of these uploaded JPG images are slightly less than 940 pixels, the DPR display system does not re-sample or re-encode the in-post JPGs which are displayed when you click on any of the images in order to expand their size when viewing them in-post. Thus, there is no need to download the Original versions from the associated DPR Gallery web-pages.
However, these smaller versions that you are now looking at (without expanding the viewing size of the images displayed in this post) have been down-sampled to a maximum pixel-dimension of 480 pixels, and have also been re-encoded to what are rather significantly reduced resolution JPGs.
.
GH2 RW2 ISO=160, Shutter Time=1/800, LGV 7-14mm, F=4.0, DxO Optics Pro 7.23 processing, mild post-USM using PSP X4 (Radius=0.5, Strength=0.6, Threshold=5):
.
Without "Lens Softness" corrections or post-USM:
.
With "Lens Softness" corrections and post-USM:
.
Without "Lens Softness" corrections or post-USM:
.
With "Lens Softness" corrections and post-USM:
.
Without "Lens Softness" corrections or post-USM:
.
With "Lens Softness" corrections and post-USM:
DM ...
Because the maximum pixel dimension of these uploaded JPG images are slightly less than 940 pixels, the DPR display system does not re-sample or re-encode the in-post JPGs which are displayed when you click on any of the images in order to expand their size when viewing them in-post. Thus, there is no need to download the Original versions from the associated DPR Gallery web-pages.
However, these smaller versions that you are now looking at (without expanding the viewing size of the images displayed in this post) have been down-sampled to a maximum pixel-dimension of 480 pixels, and have also been re-encoded to what are rather significantly reduced resolution JPGs.
.
GH2 RW2 ISO=160, Shutter Time=1/800, LGV 7-14mm, F=4.0, DxO Optics Pro 7.23 processing, mild post-USM using PSP X4 (Radius=0.5, Strength=0.6, Threshold=5):
.
Without "Lens Softness" corrections or post-USM:
.
With "Lens Softness" corrections and post-USM:
.
Without "Lens Softness" corrections or post-USM:
.
With "Lens Softness" corrections and post-USM:
.
Without "Lens Softness" corrections or post-USM:
.
With "Lens Softness" corrections and post-USM:
DM ...
Attachments
Last edited:
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
This PopPhoto.com article:
http://www.popphoto.com/gear/2012/1...ur-could-restore-blurred-and-out-focus-images
... also mentions these efforts in the field of de-blurring technology:
http://www.cse.cuhk.edu.hk/~leojia/projects/robust_deblur/index.html
Jack Hogan
Veteran Member
That's my guess also. Nevertheless they have a number of Phd's on their staff so maybe they know something I don't
I was simply suggesting that you turn off any noise reduction, sharpening or blurring that can be turned off in the raw converter, so as to have as raw an 'original' as possible - but I think reading your post that you've done that.
Yeah, isn't marketing wonderful? InFocus has three types of 'blur types' + Auto: Generic (I am assuming gaussian), Out of Focus, Straight Motion + Unknown/Estimate (Auto). I have only used Generic and Auto. Straight motion I have not been able to apply successfully (I guess when I get a motion-blurry image it's blur kernel is never a straight line) - and Out of Focus I've never really used.
As I mentioned earlier I typically like their Auto mode for capture sharpening of images where detail is important.
Yes, I had read his very informative site, but he appears to use very simple routines and uses synthetic images as examples: if the PSF is well behaved, it's much easier to get good results...
Yes, I have seen their 'magic' demo last year. Not quite ready for real-world prime time yet, but I am sure it's just a matter of time (years, not decades). Next post the GIFs
Jack Hogan
Veteran Member
Here are the GIFs, they should be pretty self explanatory. I used Topaz InFocus default pre-sharpening common preset and nothing else. The set of three includes the 'original', while the set of two just compares InFocus 'common' and DxO 'Lens Softness'. Hopefully they'll work without downloading. In any case make sure that you view them at 100% and no larger by hitting Ctrl-zero on your browser. First place to look at is the grass in the foreground, then the rest.
EDIT: The animations work in my Chrome browser if I click on the images to 'expand' them - remember CTRL-ZERO.
3 Up
2 Up
3 Up
2 Up
3 Up
2 Up
EDIT: The animations work in my Chrome browser if I click on the images to 'expand' them - remember CTRL-ZERO.
3 Up
2 Up
3 Up
2 Up
3 Up
2 Up
Last edited:
Jack Hogan
Veteran Member
I don't disagree with you, artifacts are the other side of the medal as far as having better organized information and hence IQ: it's a compromise and this particular one may not be best. Nor was the objective of the comparison to produce an artifact free image. As I mentioned earlier the 'common' preset is pushing it a bit too far for my taste and I'd prefer Auto. But then it would not match the 'strength' of Lens Softness, which to me also pushes it a bit too far, so perhaps it would not be a fair comparison.
Can you be more specific, though
1. did you make sure that you did not enlarge the image beyond 100% by hitting ctrl/cmd-zero in your browser?
2. where do you see the objectionable artifacts and are they more objectionable than having the blurrier image once you leaned a little back from the monitor?
gollywop
Veteran Member
Here, DM, are 6 versions: 3 from ACR with differing parameters and 3 using the Glenn Mitchell web script with differing level opacities. The ACR versions are named ACR x(0.5)y, where x is the amount, and y is the detail. All have radius 0.5. A detail of 100 is pure deconvolution, a detail of 0 is whatever the heck that means (nobody has given me a good answer to that question). The Glenn Mitchell versions are simply the percent opacity of the 100% effect.
I've had time only to do the "center" version.
Edit: after posting, I notice that the captions did not come through. I have found this bug in the new editor to have occurred in several of my posts.
These images are, in order:
ACR 150(0.5)0
ACR 60(0.5)50
ACR40(0.5)100
Glenn Mitchell at 65%
Glenn Mitchell at 50%
Glenn Mitchell at 35%
ACR 150(0.5)0
ACR 60(0.5)40
ACR 40(0.5)100
Glenn Mitchell at 65%
Glenn Mitchell at 50%
Glenn Mitchell at 35%
--
gollywop
I've had time only to do the "center" version.
Edit: after posting, I notice that the captions did not come through. I have found this bug in the new editor to have occurred in several of my posts.
These images are, in order:
ACR 150(0.5)0
ACR 60(0.5)50
ACR40(0.5)100
Glenn Mitchell at 65%
Glenn Mitchell at 50%
Glenn Mitchell at 35%
ACR 150(0.5)0
ACR 60(0.5)40
ACR 40(0.5)100
Glenn Mitchell at 65%
Glenn Mitchell at 50%
Glenn Mitchell at 35%
--
gollywop
Attachments
Last edited:
I was using my 3rd generation iPad, so, no, I did not press any such keystroke. The artifacts look like jpeg compression artifacts with USM halos around them. What's created is false sharpness with a disturbing character akin to the false sharpness produced by Foveon sensors, but qualitatively different. It would be much less noticeable when displayed 100% on a CRT, where blurring would mask the artifacts.
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
Hi gollywop !
Am not sure what your reference to Glenn Mitchell refers to. Can you explain what was done ?
Did some searching on the internet, and found this:
http://www.thelightsrightstudio.com/tutorials/BlendedExposuresRevisited.pdf
Is that the method which you are referring to having used on those particular images ? Please advise.
My understanding is that it is not necessarily possible to improve an image's SNR using such techiques - though such techniques are a method for implementing particular tone-curve transfer-functions. Perhaps I am wrong about. Am interested as what your understanding may be.
Detail Man
Forum Pro
- Messages
- 17,489
- Solutions
- 11
- Reaction score
- 2,745
On a recent thread (where the relevant technical posts were unfortunately for unknown reason deleted), Anders W and I discovered that the Adobe LR 3.x /4.x ( and CR 6.x / 7.x) NR (which Adobe's Eric Chan has publicly stated are the very same in both versions) works significantly better on RAW than TIF. This discovery was a rather disappointing surprise for me (as the Color NR is pretty much the only reason that I have used my copy of Lightroom 3.x Thus, it seems likely that (at least part) of their NR is done on RAW data before demosaicing.
Am suspecting that the LR/CR Sharpening (as appears to be the case with InFocus processes, as well), operates only on the post-demosaiced RGB data, though. Not sure, but that may make sense - in that the Adobe applications do not have the benefit of RAW-level sensor/lens characterization like that which DxO Labs performs in creating their Optical Corrections Modules.
Am pretty close to convinced that DxO's "Lens Softness" is operating on the RAW-level, where they can convolve the RAW-data with (some sort of separate RGB derived PSF "kernel" functions which vary in different portions of the image frame, as per results from lens/sensor characterizations).
The (supported body/lens) JPG-level DxO Lens Softness (and the Chrominance/Luminance NR for JPG and TIF) both are notably and significantly different and less effective - which implies that both are (at least in some degree) applied to RAW-level image-data before de-mosaicing.
(If) the above specualtions are true, then the resulting conundrum for RAW-processing users is that if one uses DxO Optics Pro as a "front end" in processing (in order to utilize it's efficacious optical corrections, perhaps with the exception of the Chromatic Aberration, which may fall short of RAW-level LR 4.x / CR 7.x functionality), and then exports 16-bit TIFs to LR 3.x / LR 4.x, the (somewhat preferrable to DxO's in some cases) Color NR (and the OK Luminance NR) of Lightroom does not appear to function nearly as well (it seems) as is the case when that LR/CR NR is able to be applied on RAW-level image-data. Sigh ...
One other annoyance about Lightroom (3.x, anyway), is that it has always refused to recognize or process any and all 16-bit TIFs that RAW Therapee exports. I suspect that this is no accident, and represents Adobe's spitefulness towards it's formidable freeware competitor ...
One hopeful possibility is that a newer incarnation of RAW Therapee 4.0.9.x for Win 32/64-bit is due soon (there is a major issue regarding 64-bit OS compatibility holding the release up, however). It is said to have much improved NR (which I believe that Emil Martinec has been working on, at least in part). Together with the various parameter controls available on their Richardson-Lucy deconvolution-deblurring Sharpening tool, it may (possibly) offer a more viable alternative to LR/CR (and perhaps to InFocus, as well ?). New stable releases for certain Linux operating-systems are already posted here:
http://rawtherapee.com/downloads
DM ...
Am suspecting that the LR/CR Sharpening (as appears to be the case with InFocus processes, as well), operates only on the post-demosaiced RGB data, though. Not sure, but that may make sense - in that the Adobe applications do not have the benefit of RAW-level sensor/lens characterization like that which DxO Labs performs in creating their Optical Corrections Modules.
Am pretty close to convinced that DxO's "Lens Softness" is operating on the RAW-level, where they can convolve the RAW-data with (some sort of separate RGB derived PSF "kernel" functions which vary in different portions of the image frame, as per results from lens/sensor characterizations).
The (supported body/lens) JPG-level DxO Lens Softness (and the Chrominance/Luminance NR for JPG and TIF) both are notably and significantly different and less effective - which implies that both are (at least in some degree) applied to RAW-level image-data before de-mosaicing.
(If) the above specualtions are true, then the resulting conundrum for RAW-processing users is that if one uses DxO Optics Pro as a "front end" in processing (in order to utilize it's efficacious optical corrections, perhaps with the exception of the Chromatic Aberration, which may fall short of RAW-level LR 4.x / CR 7.x functionality), and then exports 16-bit TIFs to LR 3.x / LR 4.x, the (somewhat preferrable to DxO's in some cases) Color NR (and the OK Luminance NR) of Lightroom does not appear to function nearly as well (it seems) as is the case when that LR/CR NR is able to be applied on RAW-level image-data. Sigh ...
One other annoyance about Lightroom (3.x, anyway), is that it has always refused to recognize or process any and all 16-bit TIFs that RAW Therapee exports. I suspect that this is no accident, and represents Adobe's spitefulness towards it's formidable freeware competitor ...
One hopeful possibility is that a newer incarnation of RAW Therapee 4.0.9.x for Win 32/64-bit is due soon (there is a major issue regarding 64-bit OS compatibility holding the release up, however). It is said to have much improved NR (which I believe that Emil Martinec has been working on, at least in part). Together with the various parameter controls available on their Richardson-Lucy deconvolution-deblurring Sharpening tool, it may (possibly) offer a more viable alternative to LR/CR (and perhaps to InFocus, as well ?). New stable releases for certain Linux operating-systems are already posted here:
http://rawtherapee.com/downloads
DM ...
Last edited:
gollywop
Veteran Member
Well, DM, I'm not sure what to say about Glenn Mitchell's sharpening techniques. They are available as PS actions and as scripts. They are excellent, and very involved. The problem is that they are a dying breed because Glenn has not continued to upgrade them for the latest PS versions. I believe they are still available for CS4 and CS5, but not CS6.
The only reason I was able to use his script on your image is because the WebSharpening script is the only one that continues to work with CS6. All the actions continue to work with CS6 but not the scripts. Unfortunately, the scripts are better. The actions use USM but the scripts use SmartSharpening, which is essentially deconvolution.
Glenn's techniques (separate for inkjet of various dpi and for web of various overall pixel sizes) make use of quite clever surface and edge masks along with blending options that prevent any highlight blowing from the edge boosts and level blends that prevent color shifts. Once applied their effects can be regulated by using the opacity sider on the group level.
Glenn has written a book on sharpening that is really quite a good and informative read. You can purchase a download version at
http://www.thelightsright.com/SharpeningYourPhotographsED
and you can find the scripts at
http://www.thelightsright.com/view/PhotoshopTools?page=2
The site you went to, thelightsrightstudio.com is Glenn's photo studio. The resource site is at
http://www.thelightsright.com
You can get to the full set of tools from the home page; click on the Tools tab near the top.
I'm heartbroken that Glenn is not able to keep these wonderful tools up to date. I continue to use the actions for inkjet sharpening, but prefer the script for web sharpening. I suspect in another PS iteration or two none of them will work.
Meanwhile, I consider the Ronnie Gaubert technique as my backup for web sharpening. It appears to work amazingly (surprisingly) well and is very simple to apply. I couldn't try it on your image, since I would need a version of at least 1560 pixels for the first pass before reducing it to 936 (or whatever).
Glenn also has actions/scripts for capture sharpening, but they have been supplanted by the ACR sharpening facility. I use the ACR facility for my capture sharpening but rarely use it for output sharpening, which I do in PS.
--
gollywop
Keyboard shortcuts
- f
- Forum