Does the M9 have too much noise reduction?

[scott kirkpatrick]

Advertisement (gone after registration)

I have yet to see good frame to frame comparisons on the forum in which we could see if the M9 is using smoothing to suppress luminance or chrominance noise at higher ISOs in the DNG files. There was one comparison in which a shot of a camera store employee, taken with an M8 and 35/1.4 wide open was compared with M9, same lens, same employee, both at 1250. The details on the M9 picture looked more blurred than the M8 picture, but that could be caused by all the folks waiting for their turn with the M9 jostling the photographer. I don't care what strategy Leica follows to make the best jpgs at various ISO, but I would prefer to see raw files corrected only for luminance issues, leaving noise as just one of the things to be cooked to taste during post processing.

 

So here's the worrisome piece of evidence: In the middle of Sean's M9 review, he includes a set of measurements he was asked to do by Leica engineers concerned about SDHC card writing speed. The surprise is that writing slows down at higher ISOs. The slowdown, which I assume is evidence that the DSPs are doing other things to the data, increases continuously as ISO.increases. The number of shots before the buffer fills decreases smoothly, and the time to clear the buffer by writing to the card just enough so that the camera can take its next shot is greater at 640 than at 320, greater at 2500 than at 1250. Thus it looks like there is ISO-dependent raw file processing going on at all ISOs, not just at 2500.

 

I checked whether this has been the case all along, by doing the same sort of experiment with my M8u (C mode, write DNG only, manual exposure, daylight white balance). The result is that the number of shots before the buffer fills is unchanged from ISO 160 to 1250, and decreases (from 12 to 10) at 2500. The interval between shots once the buffer is full stayed the same at all ISOs up to 1250. (The card filled up and I couldn't check this at 2500.) So it seems from this test that noise correction was only being applied at 2500, and the raw files are really raw at 1250 and below, in the M8 firmware. I think I prefer that approach, and firmware being just funky software, I think we should ask Leica to consider it.

 

scott

[Advertisement]

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

[scott kirkpatrick]

I took a careful look through Sean Reid's ISO comparison studies that he posted on 10 Sept, to see if the slowdown which starts to be measureable at ISO 640 does anything that you can see. If you look at his first series of side by side comparisons it seems to me that even at 640, the text on the WhiBal card, which is equally crisp in the 8.2 and the 9 series at ISO 160, is a bit smeared in the 9 series while remaining crisp in the 8.2 shot. Since he apparently shot each with the same lens (but he doesn't say what lens it was), this is a proper pixels to pixels comparison. And if they really are the same pixels, and the crop is from the center of the image where offset microlens effects shouldn't matter, the differences should be only due to in-camera smoothing.

 

In any raw file developer program, such as C1 or ACR, there is a switch somewhere that lets you turn all noise smoothing off. I'd like to see that offered in the M9, too.

 

scott

[ampguy]

I wondered the same as OP. Wondering if the delay and increased processing may be affecting JPGs only?

 

I took a careful look through Sean Reid's ISO comparison studies that he posted on 10 Sept, to see if the slowdown which starts to be measureable at ISO 640 does anything that you can see. If you look at his first series of side by side comparisons it seems to me that even at 640, the text on the WhiBal card, which is equally crisp in the 8.2 and the 9 series at ISO 160, is a bit smeared in the 9 series while remaining crisp in the 8.2 shot. Since he apparently shot each with the same lens (but he doesn't say what lens it was), this is a proper pixels to pixels comparison. And if they really are the same pixels, and the crop is from the center of the image where offset microlens effects shouldn't matter, the differences should be only due to in-camera smoothing.

 

In any raw file developer program, such as C1 or ACR, there is a switch somewhere that lets you turn all noise smoothing off. I'd like to see that offered in the M9, too.

 

scott

[jaapv]

Is this in jpg only? In that case it will keep the internet noise down as well.

[scott kirkpatrick]

No, the measurements that Sean and I did were with only DNGs being output. Sean describes his ISO methodology -- he shoots in DNG, processes in COne 4.8.3 with the default profile, sharpening turned off.

 

I understand that noise correction is vital for jpgs -- they're delivered fully baked. The concern is that Leica (like Canikon) is smoothing their files before we get them. That should be at most an option.

 

scott

[Jamie Roberts]

No, the measurements that Sean and I did were with only DNGs being output. Sean describes his ISO methodology -- he shoots in DNG, processes in COne 4.8.3 with the default profile, sharpening turned off.

 

I understand that noise correction is vital for jpgs -- they're delivered fully baked. The concern is that Leica (like Canikon) is smoothing their files before we get them. That should be at most an option.

 

scott

 

Scott.... I think they've done a number of things to make the high ISO better--um, usable--for the M9.

 

I'm pretty pleased with the balance between noise and detail at ISO 2500. It doesn't look smeared at all to me, and more importantly, the CCD retains its character in highlights and shadows--the Nikons / Canons start to lose DR as the ISO goes above 2000....

 

So I think they should keep tweaking it, but I'm not sure much would be gained by having a switch to turn off NR at 2500 on a CCD

[ampguy]

Advertisement (gone after registration)

Thanks Scott, this is good info. I hope the M9 gets a complete NR off option.

 

No, the measurements that Sean and I did were with only DNGs being output. Sean describes his ISO methodology -- he shoots in DNG, processes in COne 4.8.3 with the default profile, sharpening turned off.

 

I understand that noise correction is vital for jpgs -- they're delivered fully baked. The concern is that Leica (like Canikon) is smoothing their files before we get them. That should be at most an option.

 

scott

[pklein]

Scott: My guess is that they are applying NR very lightly even at 640. Looking at Sean's M9 examples, I can see little "zits" and occasional rings of high-intensity noise on what looks like a more smoothed background (look at the white test card where it is shadowed by the pile of nickels). Contrast this to the M8 shot, where the higher average noise seems to blend more naturally with the zits and rings. This looks to me exactly like what I got when I was tweaking Neat Image to remove noise without making the image "plastic." You can smooth away much (but not all) of the average noise without damaging the image, but you can't get rid of the really high-intensity stuff without turning people into Barbie dolls. My formula for NR has always been

 

1. Don't use it unless you truly must

2. Remove most of the chroma noise

3. Remove as little luminance noise as you possibly can to get the print you want.

 

I agree with you that choice is good, and an NR on-off option in the menus would be a good thing. But it does look like Leica is applying the NR very conservatively, which may satisfy most people.

 

For what it's worth, I almost never applied Neat Image to ISO 640 M8 shots. And I hardly used 1250 at all, so I don't know how it would work there. I found that Olympus E-510 shots at ISO 800 printed very nicely at 8x10-ish without any luminance noise reduction, and the E-510 is quite a bit noisier than the M8.