M9 noise reduction at medium ISO?

[noah_addis]

Advertisement (gone after registration)

I've been making some test prints at very large sizes (40-70 inches wide) and have noticed that the print quality drops dramatically when I compare photos shot at ISO 160 with those shot at 320.

 

Is the camera doing any in-camera noise reduction at medium ISO? What else might explain the rapid drop in quality between ISO 160 and 320?

 

The files at 320 have noticeably more smearing and blurring of detail, especially in complex picture areas with lots of texture and fine detail. I realize these print sizes are a real stretch for the M9, but I wanted to see how far I could push the files.

 

My process has involved various different workflows, but the best results have been to convert the uncompressed DNG files in C1, enlarge to print size at 240DPI with Alien Skin Blow-Up and in some cases add noise in Exposure 2.

 

At ISO 160 the quality is simply amazing at 27x40in. Even at 50x70in, the prints are still surpisingly good with a high level of detail and sharpness. But switching to ISO 320 the quality falls much more than I would have expected. And it's not noise, but a lack of detail that seems to be the problem.

 

I could post some crops tomorrow, but really I'm talking about prints so they may be of little value.

 

Any thoughts?

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[diogenis]

At least can you identify noise twice as much on the 320 ISO Noah? Obviously the mag is very high so noise should be easy to see.

[scott kirkpatrick]

Yes, Sean Reid's original reviews showed some evidence for noise reduction being applied well before you get to the extreme cases of ISO 1250 and above. You could see the sharpness of text on his WhiBal card degrading a bit at ISO 640 and the timing tests that he reported showed that shooting continuously slowed down gradually as ISO is increased. At the time I recall several of us thought that noise reduction that automatically goes into the RAW file ought to be optional, and I hope that Sean passed this suggestion along. Now that the firmware team probably feels that hangups and red corners are under control, and are thinking about functional enhancement, this is a good one to argue for. Do you feel like sending Stefan Daniel an example?

 

scott

[pascal_meheut]

Not only did Sean mentionned this but DxO also found that the M9 does some very light noise reduction and that this is probably the reason for the extra ISO sensitivity compared to the M8.

[thompsonkirk]

I'm glad you raised this issue, Noah, because I'd seen it in smaller prints (14x21") & wondered why. What I've seen is retention of 'main' details along with smudging of 'second order' details, the finest ones, when shifting from ISO 160 to 320.

 

I convert in LR/ACR, process in PS, & res up with Genuine Fractals. Also I print with a Canon iPF printer that sends 16-bit 600 ppi PS files directly to the printer (bypassing the usual printer driver). This print resolution/acutance might account for seeing the 160/360 difference in smaller prints.

 

I initially attributed it to using Noise Ninja on higher-ISO files, but could still see the same 'smearing' when I put NN on a separate layer & turned it off & on. So you can't blame it on the post-processing noise-reduction software. Must be the firmware, as you suggest.

 

Kirk

[Jamie Roberts]

Noah,

 

Did you see the forum topic on DNG compression? Are we only talking about uncompressed DNGs?

 

DNGs have their blackpoint set increasingly aggressively by the firmware, currently, as part of noise reduction. The blackpoint then cuts levels of information out as ISO rises.

 

If you're shooting compressed DNGs, as I understand it, you are cutting out potentially many more actual levels of detail, since the compression algorithm is already losing information.

 

This might account for some loss of detail you can see at very high enlargement levels.

[scott kirkpatrick]

Advertisement (gone after registration)

Jamie, the blackpoint shift is separate from the noise-smoothing. For the M9, in contrast to the M8, Leica seems to be trying to make everything look a little better for the average user all the time, in order to reduce the criticism that the M8 received for weak high ISO performance. This means that the black point is hiked up as ISO increases, and signal processing cycles are applied at all increased ISOs to create smoother DNG images. You can override the black point that is provided but you can never unsmooth the data to restore lost detail.

 

I think this is a slippery slope, but many would not agree with me, since the effect is small and visible mostly when you go for very high enlargement, either pixel peeping or printing. That's why I think we should ask for the chance to override the noise smoothing in the menus, for those who plan to employ post processing tools, or just prefer their pixel data the way it was in the M8.

 

scott

[Jamie Roberts]

Jamie, the blackpoint shift is separate from the noise-smoothing. For the M9, in contrast to the M8, Leica seems to be trying to make everything look a little better for the average user all the time, in order to reduce the criticism that the M8 received for weak high ISO performance. This means that the black point is hiked up as ISO increases, and signal processing cycles are applied at all increased ISOs to create smoother DNG images. You can override the black point that is provided but you can never unsmooth the data to restore lost detail.

 

I think this is a slippery slope, but many would not agree with me, since the effect is small and visible mostly when you go for very high enlargement, either pixel peeping or printing. That's why I think we should ask for the chance to override the noise smoothing in the menus, for those who plan to employ post processing tools, or just prefer their pixel data the way it was in the M8.

 

scott

 

Scott, I understand that the black point shift is different from whatever noise reduction Leica is providing; the theory is (since I've never run it across it with actual files yet) that artifacts could be introduced as a result of compression due to the black point shift... and that will lose data in addition to any NR loss.

[wattsy]

Yes, I've also noticed a difference in detail rendition between 160 and 320. I too wish there was some way of disabling the noise reduction (if that is what is going on).

[thompsonkirk]

"DNGs have their blackpoint set increasingly aggressively by the firmware, currently, as part of noise reduction. The blackpoint then cuts levels of information out as ISO rises." – Jamie

 

Twice now this thread has explained why I have problems that I was working on. I'd noticed loss of a 'Zone' of shadow detail in 320-640 files & have been working to get it back by using LR/ACR's Fill - which tends to make prints flatter in the lower midrange but without really good shadow detail.

 

(I understand that noise reduction & raising the back point are 2 different matters, but they end up making the same lower tones murky & in need of post-processing.)

 

I wish Leica would not try to solve hardware problems by tinkering with the firmware this way – the RAW files are supposed to be RAW. I understand they have to make some choices to get the recorded gray data into color, but beyond that I wish they'd quit messing with things. You can't make a better Leica with 'fixes' that move toward giving a pleasing standard point-&-shoot shot. I expect to have some processinng to do, but am upset if I'm supposed to start with a 'pushed' black point.

 

I've also noticed increased saturation, especially of reds, with the new firmware. I have to reduce this on the camera profile page of LR/ACR & have saved new Presets that do so automatically. Am I doing this to compensate for another ill-conceived 'fix'?

 

Kirk

[mjh]

I wish Leica would not try to solve hardware problems by tinkering with the firmware this way – the RAW files are supposed to be RAW.

There is some misunderstanding I think. For one thing, raw files don’t capture raw sensor data – they rarely do. Almost every camera will apply some sort of preprocessing and the M9 needs to correct for cyan corners, for example. Black-frame subtraction is also performed prior to saving the raw data. But as far as the black point is concerned, the M9 firmware is not tinkering with the raw data – rather the DNG apparently preserves the digitized values without subtracting the black level. The black point stored in the DNG is just another number; the actual image data isn’t tampered with. In fact the issue discussed here before is that the M9 doesn’t normalize the raw values to a range between 0 and 16383 by subtracting the black point (as the M8 does, prior to compressing the values to 8 bits). With uncompressed files it doesn’t really matter but with the coarser representation of tonal values in compressed files this entails a loss of information that could have been prevented by subtracting the black point.

[thompsonkirk]

Michael, maybe you missed the phrase: 'I understand they have to make some choices to get the recorded gray data into color.' And of course to do the lens corrections.

 

What's your 'take' on my main point, that M9 shadow detail is harder to retain at 320-640? It seems to me to take more post-processing to tease it out, & I find I've lately spent quite a bit of time in PS working on that. Is this a 'personal problem,' or one that others have noticed?

 

Kirk

[scott kirkpatrick]

Michael's point is that the blackpoint which is set can be overridden, and the pixel values in the raw file are not affected. (Not by the " fill " slider but by shifting the black point back down.)

 

But noise smoothing irreversably changes raw pixel values and degrades small details. It is this action which should be made optional, for those willing or needing to postprocess in software, or who just like the pixels better the old way.

 

scott

[ho_co]

Scott, Michael, Noah, Jamie, anyone--

I think I understand the technical discussion, but have a rather basic practical question:

 

Using Bridge + PSCS4, I normally begin file adjustments by clicking "Auto" in the CR "Basic" panel. I understand that "Auto" adjusts the image's black-point as well as other parameters.

 

1) Does "Auto" indeed adjust black-point as I've been told?

2) Is using "Auto" just the weak-livered lazy man's way out?

3) Could you explain to me or direct me to some material on why I should want to adjust the black-point?

4) And what would I do to adjust the black-point?

 

Thanks. Very interesting discussion, sorry to go off-topic.

[Guest WPalank]

anyone--

 

4) And what would I do to adjust the black-point?

 

Howard,

 

This is where you would adjust the black point in the CS4 ACR dialogue. There are two additional ways I know how to do it in the PS application itself.

 

 

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

[noah_addis]

For the record, I always shoot uncompressed DNG. I haven't done side-by-side testing, I'm comparing real-world results from different times and different places.

 

Before doing careful testing (I'm not much of a camera tester but I do want to know how to get the most out of my gear), I was wondering if others have had the same experiences.

 

I'm not totally sure that the smearing and smudged details I'm seeing isn't just a result of massive upsizing.

 

If it turns out to be true that there is noise reduction going on even at lower/medium ISOs, I'd definitely be in support of having an option to turn that noise reduction off.

 

Scott--I'll try, but recently I can't get anyone at Leica to talk to me about the red edge or any other M9 question. Frankly the customer service I've received is horrible. It hasn't been worthy of a much larger corporation, and certainly I would expect more of a small specialty manufacturer such as Leica.

 

With all due respect to the valuable knowledge base of this forum, I shouldn't HAVE to ask here to find out this kind of information. Leica should make this kind of information available to customers and they should return their phone calls:mad:

[Jamie Roberts]

Michael's point is that the blackpoint which is set can be overridden, and the pixel values in the raw file are not affected. (Not by the " fill " slider but by shifting the black point back down.)

 

But noise smoothing irreversably changes raw pixel values and degrades small details. It is this action which should be made optional, for those willing or needing to postprocess in software, or who just like the pixels better the old way.

 

scott

 

Scott--

 

No, Michael's (and Sandy's) point is that the blackpoint is not subtracted before compression (on compressed DNGs) and can't be "unset" from the compression scheme.

 

IOW, as Michael pointed out, it doesn't work the same way as the M8 and can contribute to more loss of detail, as I said earlier, with compressed DNGs.

 

Having said that, it won't affect uncompressed RAWs.

[jaapv]

Howard,

 

This is where you would adjust the black point in the CS4 ACR dialogue. There are two additional ways I know how to do it in the PS application itself.

 

 

[ATTACH]196377[/ATTACH]

I know how to set it in ACR, but where would one do that in C1-5 pro? I usually leave it until I get into CS4, but would prefer to set it as early as possible (sorry if this is a stupid question..)

[Advertisement]

Hallo,

Look here: Leica M

[adan]

First, I'm not sure how the "black point" crept into this thread, because what I see (or may be seeing) as evidence of aggressive noise reduction has nothing to do with shadows or blacks - it is in midtones and highlights that it is most noticeable.

 

I posted an image on (I think) the thread about the DxO review, showing what I called "bald spots" where it appears some kind of blurring has been applied, leaving the noise and fine detail smeared as though with the Photoshop "smudge" tool. Not unlike the automatic noise-reduction the Digilux 2 insisted on applying to ISO 400 images.

 

The test I did in my Photo.net review of the D2 six (really - SIX?!) years ago was to compare an ISO 400 image to an image shot with the camera set to ISO 100, but with the same exposure as the ISO 400 shot (underexposed 2 stops) - then raised to similar brightness in Photoshop. In that case the shot at ISO 100 exposed as EI 400 was noisier but also retained more fine detail compared to the straight, camera-processed ISO 400 shot.

 

I'm out to try the same test with my M9 at various ISOs (Hand-pushing an ISO 160 shot underexposed 4 stops (EI 2500) may be - um - tricky, though).

[Jamie Roberts]

I know how to set it in ACR, but where would one do that in C1-5 pro? I usually leave it until I get into CS4, but would prefer to set it as early as possible (sorry if this is a stupid question..)

 

 

Jaap--

 

The (post compression) blackpoint (for compressed DNGs) is set in C1 with the levels dialog by dragging the left-hand level indicator towards the right and into the histogram.

 

In C1 V5 you can set each channel's blackpoint individually, which--importantly--will instantly clean up many nasty colour casts (same goes for whitepoint, for what it's worth ). It's absolutely something you want to do in RAW You can always neutralize it later if you feel the need, IMO.

 

And it's not a stupid question at all... a lot of people don't know the importance of this when developing. Also, IIRC, ACR / LR does this in a totally different way algorithmically than C1, and it's one of the reasons I like C1's results better.

 

Andy--pre-compression blackpoint crept in here because it will cut out levels of detail if the RAW converter honours it. That some of that detail could theoretically fall into the lower mids or even (on a pushed file) into the upper midtones is why I mentioned it.