>would appreciate comments as to whether this microstructure is here to stay with future releases of ACR or it's an early-implementation artifact and will be fixed soon.
As far as the comments you desire, read this very long, heated, controversial thread below to gain some insight into the issue. Other than that, I doubt you'll hear any specifics, as it's Adobe policy not to comment on unreleased software.
Mike Mander, "ACR v4.1 vs v4.0: baseline high ISO NR - Part 1" #1, 2 Jul 2007 6:17 pm
Now that's the first example of microstructure that agrees with what I see as well. Especially around clouds and their wisps. Since this effect is present whether I use 4.1 sharpening or Smart sharpening, I chalked it up to general digital over sharpening. I just finished a print in which I used Smart instead of ACR, because I could go around the cloud edges with the History brush to remove the artifacts. When it came to masses of trees on a hillside, I found that the cloud edges exhibited this effect before the trees became marginal, so I could sharpen just below that point and erase the cloud stuff.
So, where did the microstructure artifacts come from in the first place?
I have some really good scans of Tech Pan film and I am going to run the unsharpened image through both and see what happens.
Ah, yes, an interesting thread :-) However it bogged down in not very fruitful discussions of what is ugly and who can see what in the final print.
The question I'm interested in is much more narrow. Does Adobe (and I guess I really mean Thomas Knoll) think that this microstructure is:
( A ) Very nice and is here to stay;
( B ) Is irrelevant and unimportant, only the concern of pixel-peepers with too much time on their hands. Nothing deliberate will be done about it;
( C ) Is an unfortunate, but acceptable side-effect of the new rendering algorithms. Nothing can be done about it at the moment and the pluses of the new algorithms outweigh the minuses;
( D ) Is a problem, should be fixed, and probably would be fixed in one of the future releases of ACR.
I have been uploading digitally captured 50mb files to two of the largest agencies in North America for 4 years, and I have rarely had any rejected.
The latest submission I sent to them, many shot on the 1Ds Mark III was nearly rejected (they did take them) but they questioned the very same microstructures/artifacting, whatever you want to call it, and accused me of using too much noise reduction. I use NO noise reduction whatsoever eccept for 25 points on the Color slider under Noise Reduction.
I have to agree that I do not like what is going on since ACR 4.1
I ran some checks using a pair of images scanned several years ago on a Heidelberg Tango. The results are that sharpening in ACR 4.1 or Smart are about the same. Setting each to match is not apparent, as Smart has no Detail or mask slider. But, in both cases, sharpening at about 70, Radius 0.6 and detail at 25 (4.1), no Mask, produced significant sharpening without the microstructure problems. Moving the Luminous slider to max changed nothing.
But, using the surface blur set at 2Px and Threshold 10 removed some garbage first, then sharpening did produce the microstructure as noted. Finally, sharpening first then Surface Blur next did show somewhat better results, but detail also suffered.
So, there it is.
Jeremy, which version of ACR are you using with the Canon 1Ds MkIII?
Sorry, that should read 1D Mark III. I am getting ahead of Canon's release date of November for the 1Ds MarkIII!
This ain't "official" but my answer would be: B & D from above...people looking at an image at 100-400% in Camera Raw 4.1 really _DON'T_ know what they are looking at but it _IS_ an issue that will be addressed.
Then, what are we looking at, Jeff? What are we looking at?
Stating "B" as one of the answers is a huge insult to those of us who are trained to look carefully.
What is "it" that will be fixed if we don't know what we are looking at?
And why counsel us to always look at 100% to be sure of what we are looking at if we really don't know?
The mind boggles....
> This ain't "official" but my answer would be: B & D from above...people looking at an image at 100-400% in Camera Raw 4.1 really _DON'T_ know what they are looking at but it _IS_ an issue that will be addressed. ( Jeff Schewe, "Weird structures in tonal transitions under ACR 4.1" #9, 10 Sep 2007 6:19 pm)
Thanks for the reply, Jeff. The fact that you accept D as at least one of the answers is encouraging :-)
Other than that I'm a bit puzzled. Surely we are expected and encouraged (by you in particular, IIRC) to look at images at 100% in ACR, especially when we're doing sharpening. I even recall somebody saying that at less than 100% magnification the sharpening effects aren't displayed at all.
It also seems fairly clear what we're looking at -- these are artifacts of the way ACR 4.1 processes raw files. Of course I don't know exactly which part of the processing is responsible for them and what specifically causes them, but from my point of view as the end user it's just details and, more importantly, details about which I can't do much.
I also take your choice of both B & D to mean "I think it's unimportant and irrelevant, but you, oh so highly annoying people, are making a fuss out of it, so we'll have to do something about it". Is that about right?
The way I see it, what Jeff is suggesting is that the new way is better in his view, but that the ACR team is looking at ways to provide an alternative choice. Pure speculation on my part.
You have to look at the image at 100% to see a preview of the sharpening controls...and ideally, adjust the controls to make the image look "properly" sharpened at 100% to recover the apparent sharpening lost in the digital process. But, looking at the image at 100% will tell you diddly squat about the resolvable details of an image printed out at 300-480PPI on a 2880 DPI printer.
So, when evaluating an image at 100%, you need to comprehend what it is you are really seeing...super high frequency noise ain't micro-detail. The edges you sharpen to enhance the image for printing is what is important.
As to Getty or some other stock agent kicking images because they don't understand what they are looking at doesn't surprise me...it's disappointing but not at all surprising.
A 72-100PPI computer display is a low resolution rendering of high resolution image data. As such, it's not at all reliable for predicting the resolvable detail of a high resolution print.
so are you saying that a high res print output will probably be sharper/show better detail then what I preview on the monitor?
I ask this question because I'm soon to be having some Giclee prints made.
Hehehe Jeff will wince at the term. :D
I believe I described what I was seeing, Jeff. I didn't call it micro detail or noise, but rather a microstructure existing at the edges of the detail, structures that cause the printed image to look unnatural. It exists as a result of Surface Blur then sharpen. Reversing the process reduces this effect. Of course, one can negate the sharpening this way, so one needs to be careful. And until I saw the OP's image, and his description, I was not crystallized in my thinking of what was happening while filtering. His post is an AHA! for me.
Predicting the final printed result is about as difficult as predicting how my piano playing will sound in a recording. IMHO, anyway. So I do not predict either. I look at the final result.
After years of playing, I can kinda predict the outcome of recording music, but printers and software change so fast that predicting what will look good now may be wrong in the future.
I don't know if there is any correlation between what I found substituting scanned for DSLR source, but at least I got a better handle on a workflow that makes more sense.
"Hehehe Jeff will wince at the term"
Well I don't have one of those high end Epson printers and here in Tucson I can't find any place that makes prints from an Epson printer. So it's either Giclee or Chromira prints. When you live in the middle of the fing desert, you take what you can get
Side note to, Ramón yes, I was kidding! (a question you asked from the thread that was closed down). But thanks for sharing
> looking at the image at 100% will tell you diddly squat about the resolvable details of an image printed out at 300-480PPI on a 2880 DPI printer.
Jeff, you're still arguing against something or somebody from another thread. Nobody here said anything about resolvable details or high-frequency noise. Besides, your statement is silly on its face -- if I'm looking at a contrasty 30-pixel-wide line, I can be pretty sure it'll resolve on the print, while a single-pixel dot will pretty sure be lost. What you probably meant is that looking at a screen it's hard to say where the boundary between the details distinguishable in a print and details lost below the perception threshold will lie -- and a 100% view won't help you with this problem. That I agree with.
> A 72-100PPI computer display is a low resolution rendering of high resolution image data. As such, it's not at all reliable for predicting the resolvable detail of a high resolution print.
A nitpick: image data doesn't have any resolution. Resolution is a function of the output device, while the image itself has only size in pixels.
But there's a bigger problem in your approach to this issue. You assume that the final output that everybody cares about is a print, a high-resolution print of a full-frame or close-to-a-full-frame image. That is certainly true for a lot of people. And it is NOT true for a lot of other people.
There's a lot of interesting stuff done with digital images, stuff that's not concerned with realistic depiction of real life, with correct colors and properly sharp details. Photoshop is a wonderfully flexible tool and its flexibility is part of what made it so successful. Being oriented purely towards conventional photography seems... unnecessarily limiting.
Adding to Kaa, even though my goal is print output, that doesn't necessarily comport to realistic photography, which is an oxymoron in any case.
I've seen these glassy microstructures too, especially in high ISO and therefore high noise) or shadow portions of high dynamic range images that you boost with shadow fill or curves adjustment. I think they are caused by oversharpening the noise combined with the new demosaic in 4.1. I can usually get rid of them by dialing down the sharpening parameters judiciously, but not always.
What I am suggesting here, conjecture at this point, is that smoothing or Blur applied before sharpening sets up the condition for such structure to occur. I would like to investigate further, and I will, but right now I am getting ready to go to Jeff's town (Chicago)for a few days.
Jeff, call for good weather through Monday, will ya? :-)
Jeff said he was heading to San Jose to "cause trouble" at Adobe. (Most welcome news.) Weather is just fantastic in the Bay Area right now.
Crossing paths. I'm going to Chicago to help my twin celebrate her birthday!
If she's your twin, that makes it your birthday too, doesn't it? Or were you two born minutes before and after midnight?
Happy birthday, Lawrence!
Yep, it's mine as well. :D
But she's getting a big surprise party, and I'm one of the surprises. Fine with me as she lost her husband a couple of years ago. My niece told me her mom wants her brother!
So I go.
Thanks, Ramon. :-)
Update for 4.2
> Windows XP, Photoshop CS3, Canon 20D, ISO 1600. The image has been converted with noise reduction luminance=50, color=50. With ACR 4.0 the sharpening was set to 100 (yes, I know, that's not the proper setting) and with ACR 4.1 Amount=100, Radius=1.0, Detail=25, Masking=0.
> This is ACR 4.1 conversion: http://www.kaax.org/pile/Microstruct/20050430_0168_microstruct_ACR41.jpg
> And this is ACR 4.0 conversion:
> These are 100% crops of a small part of the original image and show an out-of-focus edge of a dark object against a light background. Raw file is available upon request.
Now there is also ACR 4.2 conversion: http://www.kaax.org/pile/Microstruct/20050430_0168_microstruct_ACR42.jpg
and ACR 4.2 conversion with noise reduction luminance=0, color=0: http://www.kaax.org/pile/Microstruct/20050430_0168_microstruct_ACR42_NR=0.jpg
That microstructure's still there. And it's still there with all NR at zero -- it's just less visible under noise, but it's there.
Ramón, where did you get the info on post #22.
Yes, the term "giclée" is very borderline in French...
Ah, thanks, Ian! I did not catch that. I wonder what the "Schewebakka" has up his sleeves ;D
Kaa> To your original post: still in ACR 4.2, as I wrote here:
I hate that look.
Jeff, post #13> You cant' mean it!! Prints, prints, only prints, always full frame, 300dpi, 1200 dpi, no crops, no enlarging....
Where do you live? Photoshop family is sold for these purposes only? Please visit the real world, reality is so much different...
As I wrote in another forum:
I work as an art director/designer and in my workflow it is not an exception to use 100% crops for on-screen art, small part of images for prints or, on the other hand, mere 30 dpi images for large media. As a matter of fact, this is very common. Many shots are presented on screen at 100% too. I'm perfectly aware these differences wouldn't show in a 300dpi print, but unfortunately 300dpi prints of full frame images are about 20% of my output.
Every pixel counts. Even though Kaa didn't mention loss of detail, I have to... I made this simple test: I made ACR 4.0 conversion of iso 100 image, scaled it to 70% then back up to 100% and applied basic sharpening filter. Then I compared it against 4.1 conversion (layer over layer). Even then the twice resampled 4.0 was better and I would choose it right away (against the fullres 4.1). Of course it depends on individual image, but new algorithm's splotches are killing many megapixels of my camera's resolution. And I can't turn it off.
There ARE losses in detail, there ARE nasty structures (in natural textures and bokeh especially) and it IS an issue in real world situations.
Concerning the noise (which is probably the origin of all those new changes): If the noise is captured, it should be converted. If one wants, one will remove it after that in proper tab. Remember, Adobe products are Pro products. In future CS4, will we have healing clone tool only? It is so much more sophisticated then the old Clone stamp, isn't it? Will there be default noise reduction when saving psd? Or what?
Stop talking about prints only and not being important what you see at 100%, please.
4.0 quality with 4.1 functionality would be great. Adobe, don't fix what isn't broken. Thank you
Ivo, have you tried ACR 4.2?
Madmanchan> Yes, I did. The same for me. Problems even at ISO 100, far from 3+ and 4.0 quality :(
Maybe it is Canon-only problem?
One of the arguments here (though I can't believe it) is that it is nonsence to care about details and artifacts at 100% or more (!!). Then why the heck noise, being much smaller and uniform, has initiated all these core algorithm changes????
I suppose it was the noise because better detail or more natural look clearly wasn't.
Ivo and Kaa, I also see unpleasing effects.
However, they are not as dominating in my pictures, and 4.2 *has*
improved quality slightly compared to 4.1.
In that previous long thread, I posted links to DNG files, among which
there was one that I pointed out specific problems in, IMG_0180.
I've now compared the output on my Mac, Lightroom 1.1 vs. ACR 4.2 &
Photoshop Elements 4.01. I exported my LR settings to XMP, and
imported these in ACR 4.2, so that I would be comparing the same
Although what I reacted to in the first place is still present, it is
less objectionable. It's hard to tell the images apart (and therefore to see the improvements), but as the following will show, there does indeed appear to be differences that are more than mere optical illusion.
XMP file exported from LR:
4.2 import in 16 bpc sRGB, downconverted to 8 bpc, layered on top of
the LR 1.1 image, treated identically in PS Elements. Blending mode
was set to difference, layers merged and auto contrast adjustment
applied. This shows the areas changed:
There's a file that's not "-small", but even though it's a 70% quality
JPEG, it weighs in at a hefty 7.5 MB.
My temporary conclusion is that Adobe has done something in response to our discussions, but whether they've done enough isn't something I can say without comparing more imags; Kaa's example earlier in this thread appears to be over-processed in the first place.