ACR v4.1 vs v4.0: baseline high ISO NR - Part 1

Question

Hi there,
I have been participating in a couple of threads in both this forum and the Lightroom forum with regards to additional "baseline" noise-reduction that appears to be happening on high ISO Canon files. To start, here are the links to the other threads:
ACR Forum (I started that thread):
http://www.adobeforums.com/cgi-bin/webx/.3bb6a869.3bc4323e/
Lightroom Forum:
http://www.adobeforums.com/cgi-bin/webx/.3bc44a00/
Here is a summery of the issues as I see them so far:
1) I, and numerous other photographers, are objecting to what appears to be additional baseline smoothing being done to high ISO raw files in ACR v4.1, even with the luminance NR slider at zero. Most users complaining appear to be shooting with Canon.
2) At issue is the (apparent) smearing of some micro-detail due to noise-reduction and an overall artificial looking rendering of detailed scenes when examined closely. In addition, out of focus details can start looking a little posterized and indeed, I have seen that all of the above effects can be worsened by the new sharpening controls, if one is not careful in their application.
3) Not all users are unhappy with this new processing.
4) I asserted that previous versions of camera raw did not appear to do additional ISO dependent smoothing but was corrected by Thomas Knoll who indicated that they did. Regarding this, I now have confirmed that indeed he is absolutely correct (as he should be!), but the effect is far more subtle in comparison to what ACR v4.1 is doing.
6) While this might all be construed as "pixel-peeping", several of us have confirmed that the effect is also visible in larger prints. In addition, some users (myself included) are unhappy with this apparent new "direction" that Adobe is taking since many felt that in the past that ACR produced the best compromise raw conversions - lots of fine detail preserved without any *obvious* smoothing. Without the ability to disable (or minimize) this smoothing, one is now forced to live with the default levels of NR on high ISO files in ACR v4.1.
7) While I claimed that I could not see this smoothing effect in other raw files from a Fuji S5 Pro, a Nikon D2X and a Leica M8 at higher ISO's, I am now beginning to think that the "chunkiness" of the noise in some of these other cameras might simply be making the new high-ISO smoothing less effective and thus less visible than it is on Canon raw files since ISO 1600 noise it quite tight and fine on my EOS-30D and most other Canons tested.
8 ) Further to point 7, Thomas Knoll also indicated that it was only Bayer-pattern sensors that "benefited" from this new raw conversion. That is why my initial testing did not show the Fuji S5 Pro (which has a complex hexagonal array Super CCD and not a simple rectangular array sensor) to suffer these same effects.
9) I presume this ACR comparison to hold true for Lightroom v1.0 compared to v1.1 as well.
In any case, what I have done is take a several ISO 1600 images from my EOS-30D and one ISO 1600 image from a D2X and make two versions of each image. One is the original CR2 or NEF file and the second, I used "exifedit" softare to modify the EXIF "iso-speed" field so that it no longer reads 1600, but rather 100. Opening these two versions of each raw file will show if there is any additional high-ISO-dependent differences in the raw conversions.
Indeed both ACR v4.0 and ACR v4.1 show a greater degree of noise reduction on the original ISO 1600 versions than when they are "tricked" into thinking the raw was shot at ISO 100. The effect is much greater in ACR v4.1 however and was immediately apparent to me within seconds of opening my first high ISO raw file. However with previous versions of ACR, it was so subtle that I never noticed it at all.
[...continued, with samples in next posting]

Answer 1

[Part 2 ...continued from previous post]
I have assembled several PSD files, each with 4 layers labeled as such (with description following the -->):
ACR v4.1 - ISO 1600 as 1600 --> original ISO 1600 raw file converted with ACR v4.1
ACR v4.1 - ISO 1600 as 100 --> same raw file but with EXIF edited to read ISO 100, converted with ACR v4.1
ACR v4.0 - ISO 1600 as 1600 --> original ISO 1600 raw file converted with ACR v4.0
ACR v4.0 - ISO 1600 as 100 --> raw file with EXIF edited to read ISO 100, converted with ACR v4.0
In both versions of ACR, the luminance NR slider was at zero, the chroma at the default of 25 and all the sharpening off or minimized. I have not found the chroma NR slider to have any visible impact to actual image detail, at least not in any of these samples, so I left it at 25. White balance was As-Shot and other controls at default/zero. No additional processing or sharpening was applied to any of the samples, apart from cropping the image to make it smaller for download.
Here are the samples:
http://www.beauphoto.com/galleries/images/D2X_Blur_Compare_crop.psd.zip
(5.39 Mb) An ISO 1600 D2X raw comparison. I use the word "blur" since we are looking at an out-of-focus background portion of the shot. Sadly the shot has no truly sharp detail to compare.
http://www.beauphoto.com/galleries/images/30D_Blur_Compare_crop.psd.zip
(6.06 Mb) An ISO 1600 30D raw comparison. Included to have a similar image to compare to the Nikon.
http://www.beauphoto.com/galleries/images/30D_Outdoor_Compare_crop.psd.zip
(13.14 Mb) An ISO 1600 30D raw comparison. The outdoor shot I had linked to in my other posts.
http://www.beauphoto.com/galleries/images/30D_Statue_Compare_crop.psd.zip
(5.79 Mb) An ISO 1600 30D raw comparison. The indoor Las Vegas statue shot I had linked to in my other posts.
The above files are just crops from the original full images. Decompress the zip and open them up in Photoshop, either at 100% or 200%. Simply check and uncheck each layer as needed to make comparisons between the different versions. When comparing, you might try to apply a small amount of sharpening (one pass of "Sharpen" for example) equally to each layer as the differences might become more apparent. Also, people viewing the samples on a CRT might find the differences less obvious than those using an LCD display.
My conclusions are as follows:
1) I believe that ACR v4.1 is indeed doing additional "baseline" smoothing of high ISO files. This is most apparent on the Canon examples but the noise is so coarse on the D2X file that is hard to be certain whether (a) there is no additional smoothing on the high ISO version, or (b) whether the smoothing is of insufficient strength (or radius?) to be clearly visible.
2) I personally believe there to be a slight reduction in "micro-detail" when using "ACR v4.1 - ISO 1600 as 1600". On some images more so than on others. In addition, there is a "smoothed" and "edge-detected" look to small details as well as some slight "furriness" where the NR algorithm appears to be deciding where edges of detail are and where smoothing should occur.
3) More to point #2, I prefer all three other versions of the converted files. That is, to my eye, the worst looking ones are "ACR v4.1 - ISO 1600 as 1600", at least on the Canon files.
4) However the very best version, when all aspects of image quality are being considered, is the "ACR v4.1 - ISO 1600 as 100" version! The ever-so-slight reduction in overall noise does not appear to affect micro-detail to any significant extent, yet there is a significant reduction in "white-specks" in darker areas (especially after post-sharpening) and certain areas of color in the images are identified and rendered better - the last item being really hard to describe. Look at the dark engraved lines on the head of the statue: ACR v4.0 fills them in as slightly "reddish" whereas in ACR v4.1 they are black and look more "normal" to me. On the outside shot, looking and the red brick colour of the wall in the distance (the one with all the patios on the right), ACR v4.1 seems to better identify where the colour should be, whereas ACR v4.0 slightly desaturates portions of that wall between floors.
5) Further to point 4 is that indeed there is obviously some very complex processing going on, and if one could reduce the level of NR on an ISO 1600 file, down to the level which ACR uses when fooled into thinking it is an ISO 100 file, I would certainly be quite happy with the results. I imagine many others would be too...
6) Finally, this effect will certainly not be visible in smaller prints however the more you push a file (for example a 20x30 print from a 30D will have a source resolution of about 117dpi) or if you end up cropping an image, thereby increasing the overall magnification and thus also the defects, well then those processing artifacts can become visible upon close and critical inspection - large prints of landscapes for example.
My humble request:
In ACR v4.2 or LR v1.2, please have the luminance NR slider, when set to zero, have the same degree of effect as when an ISO 1600 file is processed as though it were ISO 100. Alternatively, put in a "Preserve all texture detail" checkbox in the detail section that does the same thing. That way we'll have the best of both worlds: for some types of images, one can use this new-found intelligent NR but for those where the most natural look and finest detail is of utmost importance, we can turn it off - or at least reduce it to more-or-less the same level as previous versions of ACR.
In general, the new NR processing does seem quite "intelligent" and is a big step up from the previous version's crude luminance slider. I certainly do not want to change back to the old system as there are many real and visible benefits to the processing in the new ACR! In addition, the new sharpening controls are also a big step up from previous version, offering much more fine control and less artifacting, when used judiciously.
However, please leave the majority of sharpening and noise-reduction decisions to the individual user. That way, one can selectively apply these effects in layers or by using the history brush. I would be equally unhappy if suddenly there was an aggressive new baseline sharpening applied to all raw conversions, when the sharpening controls were at zero!
I realize now that maybe there never was a true "zero" in NR and sharpening, but with previous versions of ACR, this processing appeared to be very subtle and lent itself very well to post-NR and post-sharpening. This is why, even after trying out virtually every other raw converter made, I always came back to Adobe Camera Raw, time and time again.
Please don't give up on progress and on new and innovative ways to improve ACR, but also please don't take away those qualities that have made a legion of photographers use Adobe Camera Raw as their raw converter of choice!
Best Regards,
Mike Mander
http://www.sublimephoto.com