Ad
Ad
Ad
Pages: [1] 2 »   Bottom of Page
Print
Author Topic: Canon G10 RAW: ACR 5.2 vs. DPP 3.5  (Read 18863 times)
Slobodan Blagojevic
Sr. Member
****
Online Online

Posts: 6031


When everybody thinks the same... nobody thinks.


WWW
« on: November 30, 2008, 08:36:08 PM »
ReplyReply

This is a continuation of sort of an existing thread [a href=\'index.php?showtopic=28859\'](here)[/a]. However, since it was started more than a month ago, and was initially about raw converter availability for Canon G10 files, I thought it makes sense to start a new thread about a specific issue related to now available Adobe Camera Raw 5.2, namely its noise treatment.

Over the years I settled on ACR (and Lightroom, as of version 2) as my main conversion engine. I always had Canon DPP and would try it from time to time, mainly for some linear conversions, but since its functionality and interface leave a lot to be desired, it has never become my converter of choice. However, given that until very recently it was the only option for G10, I did develop my first files with it. Then came along ACR 5.2. Having had the same file already converted in DPP, I was surprised to see ACR resulting in a much noisier image. I do understand that "much" noisier is quite subjective, so I will attach some comparison shots for you to decide.

The original shot was at ISO 800, slightly underexposed (exp. adjustment was -2/3) in order to prevent highlights from blowing out. And before you see the grotesquely enlarged grain in the comparison shots at 200%, I want to point out that this post is not about how G10 is noisy... it is not. Even this slightly underexposed shot would print as a decent 8x10,  without any external noise-removal software. This is strictly about differences between ACR and DPP.

OK.. to the comparison finally. There are several similar comparisons on the web, notably DPReview of G10 (Section 12: Software & Raw Mode), which typically show two sets of comparisons: one with both noise controls and sharpening at zero, and the second with noise controls at maximum (and sharpening still at zero). While still indicative of differences, such a comparison actually masks one type of noise with another. Hence my first comparison here has chroma noise reduction at maximum, and luma noise reduction at zero (for both files a Neutral Picture Style was used, as any other would introduce even more saturation to chroma noise) . My understanding is, and correct me if I am wrong, that way we are looking at luma noise only. This is a 100% crop, and before you judge the G10 noise harshly, remember that we are looking at an almost 3x4 feet enlargement:

[attachment=10010:Picture_5.png]

I hope you can see the difference yourself in the above picture, but is is even more pronounced at 200%:


[attachment=10011:Picture_3.png]

I see here two things: 1. Even with chroma reduction at maximum, ACR still has nasty color blotches visible and 2. DPP noise is much tighter and smaller, almost like what we film dinosaurs used to know as "finer grained", hence creating an impression of a sharper picture.

I have three other combinations at 200% available (i.e., noise zero, noise max, and luma only at max) and will post them in replies if necessary (trying to keep the size of this post manageable).

As I said above, I am a long-time ACR user, hence no agenda here, other than curiosity.

Your comments are welcome and thanks for reading this far :-)

Slobodan



Logged

Slobodan

Flickr
500px
Schewe
Sr. Member
****
Offline Offline

Posts: 5497


WWW
« Reply #1 on: November 30, 2008, 11:30:06 PM »
ReplyReply

Quote from: slobodan56
Your comments are welcome and thanks for reading this far :-)


What settings EXACTLY did you use in both apps? Hard to draw any conclusions as it is...
Logged
Slobodan Blagojevic
Sr. Member
****
Online Online

Posts: 6031


When everybody thinks the same... nobody thinks.


WWW
« Reply #2 on: December 01, 2008, 12:11:11 AM »
ReplyReply

Quote from: Schewe
What settings EXACTLY did you use in both apps? Hard to draw any conclusions as it is...

Thanks for taking the time to read it through.

In ACR 5.2: Opened the file and left everything on default (i.e., no auto adjustment), except zeroing sharpening and luma noise reduction, while maximizing chroma reduction to 100. In Camera Profiles, I chose "Camera Neutral" (choosing "Adobe Standard" results in even more pronounced chroma noise). That's it.

In DPP 3.5: Opened the file and left everything on default, except zeroing sharpening and luma noise reduction, while maximizing chroma reduction to 20. In Picture Style, I chose "Neutral". That's it.

My color space in ACR is ProPhoto RGB, 16 bit. In DPP it is Wide Gamut RGB.


Logged

Slobodan

Flickr
500px
sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #3 on: December 01, 2008, 06:40:28 AM »
ReplyReply

Looks to me like ACR is doing a lot more Chroma noise reduction than DPP; I suspect that if you dialed back ACR's chroma noise reduction, you'd get more comparable results.

Sandy
Logged
madmanchan
Sr. Member
****
Offline Offline

Posts: 2110


« Reply #4 on: December 01, 2008, 07:57:45 AM »
ReplyReply

Slobodan, yes, I would say your description of the residual noise characteristics is fair. Essentially the residual noise distributions are different from the two converters.
Logged

Slobodan Blagojevic
Sr. Member
****
Online Online

Posts: 6031


When everybody thinks the same... nobody thinks.


WWW
« Reply #5 on: December 01, 2008, 10:46:55 AM »
ReplyReply

Quote from: sandymc
Looks to me like ACR is doing a lot more Chroma noise reduction than DPP...

You mean the other way around? In my examples, ACR is on the left, and DPP on the right.

To me, the picture on the left (ACR) still contains chroma (color) noise, in spite of the chroma slider being at the maximum. To me, DPP does a better job in eliminating chroma noise, and it does a better job in reading the raw file when it comes to luminosity noise (i.e., with zero luma noise reduction applied, it generates a "finer grained", film-like luma noise. It appears that a DPP file, with no chroma noise, and smaller, tighter, more uniform luma noise is a much better candidate for a specialized noise-removal software (or even for a luma-noise reduction in DPP).

In contrast, the remaining color blotches left after ACR's maximum chroma-noise reduction are next to impossible to remove in a specialized noise-removal software. The only way I can think of removing the residual chroma noise in an ACR-generated file is blurring the a* and b* channels in LAB mode in Photoshop.
Logged

Slobodan

Flickr
500px
oldcsar
Full Member
***
Offline Offline

Posts: 126


« Reply #6 on: December 02, 2008, 01:52:09 AM »
ReplyReply

It seems to me that DPP is using some aggressive chroma noise reduction, even with its NR set to zero. Based on your samples, I would say that the DPP sample is exhibiting noticeable desaturation when compared to ACR... what catches my eye the most is some desaturated blues and magentas, which may be connected to the automatic chroma/fringe reduction feature of DPP. Have you tried the Targeted Adjustment Tool for desaturating the bluish blobs of chroma noise?

Observe the blue fringing in the reflection of the upper eye in ACR and the suppression of it in DPP... could there possibly be some consequences to the baseline noise reduction that DPP is applying, at the expense of color accuracy? Chroma noise may be an example of false color, but is it possible that the aggressive treatment to fix the colored blob effect of tiny, noisy sensors results in color inaccuracies which make it worthless to push NR so far? While luminance NR is arguably similar or even incrementally improving with these tiny sony sensors (I previously owned an S40 and S50, and currently own a G9) at the print size level, I might hazard to theorize that chroma noise is getting blotchier. Let's be realistic here... one can't expect a clean sensor signal as it is with these cameras at ISO 800, and pushing NR to remove blue chroma noise MUST have some sort of perceptible impact on the overrall color accuracy if so much of the file itself is covered in chroma noise. To my eyes, the blue blobs are still visible as patterns in the DPP sample, albeit greatly suppressed (but again, possibly at expense of color accuracy). Believe me, I would love it if Camera RAW/LR could better handle blotchy chroma from my G9 at high ISO, but is it truly possible without the expense of color accuracy?

As far as the luminance NR that appears to express itself in the ACR conversion, I could be mistaken but this is the result of noise reduction during the demosaic process. Looking very carefully at the DPP shot and back at the ACR version, I would say that even at these high magnifications, the ACR version is not exhibiting a noticeable loss of detail. Compare the detail of the eyelashes as well as the corners of the eyelid... all meaningful detail appears to be retained. The only thing which is lacking in the ACR sample is a uniform layer of fine noise, which does absolutely nothing but fool our eyes into believing that it contains detail. This is why specialized interpolation programs, such as Photozoom Pro, offer Artificial detail sliders to add noise to the final output--- it fools our eyes into believing that the image is sharper and contains micro-detail upon close inspection.

Actually, I prefer ACR output as far as the demosaiced results go, compared to DPP. I personally would prefer to have some sort of noise reduction at the demosaic level... this makes the files stand up better to capture sharpening (for instance, you are sharpening less luminance noise and more actual detail). However, I would not say that ACR baseline noise is by any means perfect. So, in response to the belief that DPP files make better enlargements, I'm not so sure about that. I think that adding artificial film grain as a final step in the interpolation process would more than compensate what misgivings you have about a poster-size representation of the subject. Then again, you might be disappointed anyway if you try to make a great poster sized print from an ISO 800 G9 file in less-than-optimal lighting.



Logged

sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #7 on: December 02, 2008, 11:36:26 AM »
ReplyReply

Quote from: slobodan56
You mean the other way around? In my examples, ACR is on the left, and DPP on the right.

No. Have you tried reducing noise reduction setting? ACR does chroma noise reduction in somewhat counterintuitive way.....

Sandy
Logged
jdemott
Sr. Member
****
Offline Offline

Posts: 434


« Reply #8 on: December 02, 2008, 12:44:49 PM »
ReplyReply

Hello Slobodan,  Like you, I was using DPP with the G10 while waiting for ACR to be updated.  Most of my shots have been at ISO 80, but after reading your post I decided to look at ISO 800 comparisons of the two converters. I used Adobe RGB color space for both ACR and DPP and used the Camera Standard setting for both.  I accepted the default settings from both conversions except I initially turned off all sharpening and noise reduction.  Viewed at 200 percent, the appearance is similar to what you describe--DPP produces a finer grained appearance (albeit with more noticeable cross hatching artifacts) while ACR produces a blotchier appearance with more pronounced color noise.  Since all noise reduction is off, I assume this is the result of different de-mosaicing routines (perhaps the DPP interpolates using more distant pixels?) but I don't have enough knowledge of de-mosaicing to make an informed judgment.  

As you note, if maximum color noise reduction is applied in both converters, the effect is more pronounced in DPP.  I don't attribute much significance to that as I don't use the raw converter as a primary tool for noise reduction (I use Noiseware Pro plug-in) and I don't see any reason why the maximum NR setting on one converter should correspond to maximum on another (but zero NR should mean zero NR on both).  Again, I would guess that DPP is using more distant pixels to average the color information as a method of noise reduction.  In any event, as you say, after applying maximum color NR the DPP file is much cleaner and the ACR file still shows significant blobs or blotches of color.

Opening files from both converters into Photoshop (without NR or sharpening) and applying Noiseware Pro at default settings greatly improves both files and I find the results far preferable to using noise reduction from either converter.  The DPP file still shows noticeably less color noise than the ACR file, but it also appears to have less color depth or saturation and less tonal contrast than the ACR file.  Viewed at 50 percent and normal viewing distance, the better contrast and color of the ACR file are subjectively more pleasing to me.  The additional saturation and contrast from ACR are a little surprising to me since my general impression from working with ISO 80 files is that DPP produces a slightly more contrasty, more vibrant result than ACR at default settings.

As a general observation, it is evident that small changes in the settings for saturation, contrast, etc., can make very noticeable differences, both in the overall appearance of the shot and when examining fine detail at high magnification.  Choosing default settings may be an "objective" way to compare two raw converters, but we need to remember that the reason the adjustments are available to us is so we can use them.  If the goal is to produce the best print that each converter can generate, I doubt one would choose to stick solely with the default settings for either converter.  Finally, although it is possible to get a usable file from a low light, 800 ISO shot on the G10, the results are far short of the exceptional quality seen in ISO 80 shots.
Logged

John DeMott
Slobodan Blagojevic
Sr. Member
****
Online Online

Posts: 6031


When everybody thinks the same... nobody thinks.


WWW
« Reply #9 on: December 02, 2008, 03:37:21 PM »
ReplyReply

Quote from: sandymc
... ACR does chroma noise reduction in somewhat counterintuitive way.....
And God moves in mysterious ways too... but I doubt ACR is related.   I would welcome, however, your elaboration on exactly how counterintuitive it is.

But in all seriousness, I am puzzled by your comments. It seems you are implying that using less chroma noise reduction would result in less chroma noise? You also seem to imply, contrary to my observations, and contrary to some replies here, and contrary to a lot of other postings on the web, that ACR does a better job in chroma noise reduction than DPP?

P.S. It seems the discussion has turned to chroma noise here, although the purpose of the examples attached was to separate the two types of noise, and look only at the luminosity noise (by maximizing chroma noise reduction, i.e., minimizing chroma noise).

Once again, parameters of the posted examples:

Sharpening: ZERO
Luma noise reduction: ZERO
Chroma noise reduction: MAX

Logged

Slobodan

Flickr
500px
DarkPenguin
Guest
« Reply #10 on: December 02, 2008, 04:06:40 PM »
ReplyReply

I do not find this surprising.  I think ACR's noise reduction is okay for DSLR's relatively light noise (although some of the Googleplex pixeled cameras might prove this wrong) but it is under powered for heavy noise reduction.

Edit: Added the word "but".
« Last Edit: December 02, 2008, 04:07:25 PM by DarkPenguin » Logged
Slobodan Blagojevic
Sr. Member
****
Online Online

Posts: 6031


When everybody thinks the same... nobody thinks.


WWW
« Reply #11 on: December 02, 2008, 07:10:02 PM »
ReplyReply

Thanks for the detailed and thoughtful comments.

Quote from: oldcsar
... DPP sample is exhibiting noticeable desaturation when compared to ACR... some desaturated blues and magentas, which may be connected to the automatic chroma/fringe reduction feature of DPP... could there possibly be some consequences to the baseline noise reduction that DPP is applying, at the expense of color accuracy?...

If by "the automatic chroma/fringe reduction feature of DPP" you are referring to "Lens aberration correction", and within it "Chromatic aberration" and "Color blur" features, it is worth noting that the feature is grayed out for G10. Even for cameras/lenses it is designed to work with, it does not do anything unless you press the "Tune..." button.

As for the desaturation/color accuracy comments, I have not noticed any "noticeable" desaturation, but that is subjective, of course. I am therefore attaching another screen shot, so that you can see for yourself. This is the same shot as before, except it comes from a corner of the picture (and is out of focus), with a relatively saturated blue color to begin with. Once again, luma noise reduction is at zero, and chroma reduction at maximum:

[attachment=10074:Picture_1.jpg]

What I see here is that there is no significant difference in saturation between two raw processors. There is a slight difference in how each processor interprets colors of course, but there isn't any noticeable (or better to say: significant) desaturation as a consequence of the maximum chroma noise reduction. I used "Neutral" setting for both processors, hence the appearance of a lack of saturation that you might have referred to.

Since one can argue that it is not clear from the above example how chroma noise reduction affects saturation, here is the shot where all noise reduction is zeroed:

[attachment=10075:Picture_2.jpg]

Quote
... The only thing which is lacking in the ACR sample is a uniform layer of fine noise, which does absolutely nothing but fool our eyes into believing that it contains detail. This is why specialized interpolation programs, such as Photozoom Pro, offer Artificial detail sliders to add noise to the final output--- it fools our eyes into believing that the image is sharper and contains micro-detail upon close inspection.
Exactly! But that is the whole point, isn't it? That is why we use unsharp masking too, to fool our eyes, don't we? The way you put it, looks like you think it is a bad thing that DPP generates a file with a "uniform layer of fine noise", as opposed to ACR.

Once again, I appreciate your time and effort.

Logged

Slobodan

Flickr
500px
madmanchan
Sr. Member
****
Offline Offline

Posts: 2110


« Reply #12 on: December 02, 2008, 08:57:11 PM »
ReplyReply

Quote
but zero NR should mean zero NR on both

This is incorrect. DPP applies fairly significant NR even when NR is set to zero.

(In general, it is not safe to assume that just because a slider is zero, it means that no processing is being done.)
Logged

Ray
Sr. Member
****
Offline Offline

Posts: 8900


« Reply #13 on: December 02, 2008, 09:39:17 PM »
ReplyReply

I too have noticed these subtle differences between DPP conversions and ACR conversions which, with all noise reduction off, can be characterised as a finer grain for the DPP conversion and a slightly blotchy appearance for the ACR conversion.

However, I'm not convinced that such small differences amount to anything in the context of the general processing and massaging that one usually applies to an image to get it ready for display or printing.
Logged
oldcsar
Full Member
***
Offline Offline

Posts: 126


« Reply #14 on: December 03, 2008, 05:42:24 PM »
ReplyReply

Thanks for the additional crops Slobodan, as these more accurately illustrate the issues at hand. The reason I bring up the issue of fringing/chromatic aberration reduction is that in ACR, the full edge removal option noticeably desaturates the colors in the image. If a comparable correction is taking place with DPP conversions, then that would explain why the blue chroma noise is less expressed in the DPP 100% crops. The new crops you have provided, as well as your higher expertise in DPP, suggest that my theory is false in this respect.

However, as Madmanchan points out, there is some chroma noise reduction taking place in DPP even when the NR slider is set to zero. I do not at all disagree with you that DPP creates more pleasing 100% crops, but there is still the uncertainty of whether or not the fine-grained noise has a real competitive advantage over ACR's results (given that the noise may be invisible in normal sized prints, and in larger prints there is the possibility of creating finer noise than the interpolated pixels by adding a noise filter in post-production). I totally agree with you in principle on the perceptual value of a certain amount of noise, but in the case of large prints I would not simply decrease the PPI to create a larger print and let the printer handle the rest. I realise this is a recent trend in print production, and although I don't disagree with its practice, it has disadvantages, because I see value in creating much finer noise when interpolating pixels in the final output. What I am saying is that there is possibly an advantage to ACR's implementation of its baked-in luminance NR, but on the other hand I really cannot defend the performance of ACR's chroma noise treatment with the G9/G10. My *limited* experience with DPP has suggested to me that the images have a lower threshold for sharpening with this fine smattering of noise, resulting in a "crunchy" image far sooner than ACR.

As for your belief that ACR's method of baked-in NR reduces the effectiveness of 3rd party NR programs, such as Noise Ninja or Noiseware, if you could substantiate this as well through crops, the content of your post may be a useful suggestion for improvement on the Adobe.com forum. Since many seem to agree that its chroma NR is underpowered when compared to DPP (with G9/G10), then the other option is third party chroma NR... but if ACR's own conversions hinder 3rd party effectiveness, then I would consider this a serious competitive disadvantage for Adobe, given that DPP is a free product. But to repeat myself on the issue of luminance NR (set to zero in ACR), I do not think it is a real problem.

thanks for your insights,
Brendan
« Last Edit: December 03, 2008, 05:46:04 PM by oldcsar » Logged

sandymc
Sr. Member
****
Offline Offline

Posts: 269


« Reply #15 on: December 06, 2008, 06:27:48 AM »
ReplyReply

Quote from: slobodan56
But in all seriousness, I am puzzled by your comments. It seems you are implying that using less chroma noise reduction would result in less chroma noise? You also seem to imply, contrary to my observations, and contrary to some replies here, and contrary to a lot of other postings on the web, that ACR does a better job in chroma noise reduction than DPP?

ACR is different in that sometimes it tends to behave is if it were interpolating in a Lab (or similar) space, rather than an RGB space, and one possible reason for what you seeing is that ACR is effectively trying to interpolate chroma noise.

But rather than debate the theory, just try it. I'm very happy to be wrong - mostly, I am(!)

And no, I'm not saying ACR is "better", just that it's different, and may benefit from different settings.

Sandy
Logged
bjanes
Sr. Member
****
Offline Offline

Posts: 2822



« Reply #16 on: December 06, 2008, 08:30:29 AM »
ReplyReply

Quote from: oldcsar
Thanks for the additional crops Slobodan, as these more accurately illustrate the issues at hand.
Brendan

As Jeff Schewe and others have pointed out, examination of images at high magnification on the screen can be misleading. Often noise is more prominent on the screen than in large prints. Sharpening and noise reduction are basically antagonistic processes and one has to try to achieve an optimum balance between the two processes. The ultimate results are best evaluated in the print. Since so many variables are involved and making prints is expensive and time consuming, detailed testing is difficult.

Bill
Logged
Ray
Sr. Member
****
Offline Offline

Posts: 8900


« Reply #17 on: December 06, 2008, 06:27:29 PM »
ReplyReply

Quote from: bjanes
As Jeff Schewe and others have pointed out, examination of images at high magnification on the screen can be misleading. Often noise is more prominent on the screen than in large prints. Sharpening and noise reduction are basically antagonistic processes and one has to try to achieve an optimum balance between the two processes. The ultimate results are best evaluated in the print. Since so many variables are involved and making prints is expensive and time consuming, detailed testing is difficult.

Bill

That is exactly so. It's just so easy to compare images at 100%, 200% and upwards on one's computer monitor and get some satisfaction from the fact that differences in noise, resolution etc are clearly visible.

However, the practical significance of such differences is another matter. I generally get on the print pretty much what I see on my monitor, so I guess on the rare occasions I might want to make a large print from a very small crop of a 10, 15, 21mp image, then such differences between DPP and ACR conversions might be of concern.
Logged
Dale_Cotton2
Jr. Member
**
Offline Offline

Posts: 94


WWW
« Reply #18 on: December 15, 2008, 09:13:08 AM »
ReplyReply

I'm in the process of looking for a low light DMD so the issue of raw noise handling is especially important to me. I've been using Lightroom exclusively for my conversions since the cruel death of RawShooter, but slobodan56's analysis in this thread caused me to re-examine that. Since I don't have access to DPP, I downloaded Raw Therapee and converted high ISO files from multiple cameras in both Lightroom (in my case, version 1.4) and Raw Therapee. The difference is dramatic. By converting in Raw Therapee with zero NR, then running the resulting TIFF through Noise Ninja, I'm gaining at least a one stop advantage over the same procedure done with Lightroom's output.

Where in Raw Therapee noise is confined to single pixels, in Lightroom that pixel's immediate neighbours are falsely coloured. This suggests to me that during demosaicing Lightroom grabs the colour value from each neighbouring pixel regardless of whether it's a noise pixel or not, while Raw Therapee puts a noise pixel in quarantine. Raw Therapee and, presumably DPP, may do this, as oldscar suggests, by evaluating a broader neighbourhood.

I applaud Eric's continuing candidness and balance in communicating such issues. Since I work almost entirely at high ISOs, I hope my exile from Lightroom will be a short one. But as a programmer myself - and always assuming my armchair analysis is anywhere in the right ballpark - I can see that demosaicing has to be a core-level module and any tweak to it, will have to be thoroughly evaluated. One solution might be to present the tweak as a non-default option that users can turn on at their own risk.
Logged
Tim Lookingbill
Sr. Member
****
Offline Offline

Posts: 1201



WWW
« Reply #19 on: December 27, 2008, 05:57:44 PM »
ReplyReply

If you photographers don't get off this chroma noise issue viewing digital captures at zoom levels that will never be seen on a print, you're going to destroy the richness and depth in your images by the time ink hits paper.

Coming from a painting background I can tell you the ACR version will have a slight edge in better representing depth on a 2D surface over the DPP version. I have to admit I wish ACR would adopt a more finer dithered pattern in luminance noise reduction but the majority of the chroma noise should be left in to prevent the very life and vibrance from being sucked out of it on the print.

Here's some links that explains what I'm getting at:

http://painting.about.com/od/colourtheory/...or_theory_8.htm

http://barbarajcarter.wordpress.com/category/color/

And this one on printmaking...scroll down and click on illust. #2

http://www.rochester.edu/College/ENG/blake...enhanced/2.html

Learn from the masters. They knew so much more about representing reality on a 2D surface than any camera manufacturer.

Just my 2.
« Last Edit: December 27, 2008, 05:58:55 PM by tlooknbill » Logged
Pages: [1] 2 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad