M9 compressed DNGs, blackpoint, and noise levels

[Jamie Roberts]

Advertisement (gone after registration)

Well, I honestly love my M9--even with the firmware upgrade. I want another one

 

Now that I actually have files to hand, I have to say it's everything I wanted... So here's a question for all of you, (and maybe most of all for Sandy)

 

Did we ever figure out what Leica is doing to the blackpoint in compressed DNGs? And more importantly what impact that has on processing (in C1 for my interests?) I searched and found the original thread, but I'm wondering if anyone has done any tests (I know Jaap is finding ISO 2500 JPEGs better than RAW files in terms of noise, and I had some conjecture about black point interpretation, but I don't really know)...

 

Any info out there? Thanks in advance.

 

I do have to say practically that coupled with the latest C1 I could use ISO 2500 on the M9 in "normal" dark exposures, which is outstanding. I will post more fully on that subject later

[Advertisement]

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

[jaapv]

Yes- but following your advice I found that the noise performance of RAW could indeed be improved - to about the same level as JPG, but with better color rendition- by using compressed DNGs. So I feel you have got hold of something.

[sandymc]

Jamie,

 

I think our understanding is still pretty much what it was at the time of that thread, which I would summarize as follows:

 

1. The black point in all M9 DNGs changes (increases) with ISO

 

2. That has effectively no impact on uncompressed files.

 

3. For compressed files however it does have a negative effect in that you effectively "lose" some of the 256 codes that the M9 compression scheme allows, because those codes are then used to encode below-black point data, not real image data. The number of codes "lost" increases with ISO, because of the increase in black point with ISO.

 

I think that in practical use very few people have found this to be an issue.

 

I one thing I would add is that I haven't looked at a compressed M9 image taken with the new firmware to see whether this is still the case - If someone wants to post one I'll take a look.

 

Regards,

 

Sandy

[sandymc]

For those that might think that Jaap and I are saying totally contradictory things, btw, we're not. Fewer available levels in an image will give lower visible noise. Simplistically, the further apart the levels are, the less chance that a given amount of sensor noise will result in a pixel value changing.

 

However, before you believe that fewer levels are a good thing in high ISO images, they're really not as (a) there is the danger of banding, and ( while the image with lower fewer levels will look better in its "raw" state, the image with more levels will generally look better after being passed through noise reduction software, as there's more information to work with.

 

Sandy

[scott kirkpatrick]

Hi Sandy. I'll Yousendit one to you with compressed DNG and hi ISO.

 

scott

[jaapv]

For those that might think that Jaap and I are saying totally contradictory things, btw, we're not. Fewer available levels in an image will give lower visible noise. Simplistically, the further apart the levels are, the less chance that a given amount of sensor noise will result in a pixel value changing.

 

However, before you believe that fewer levels are a good thing in high ISO images, they're really not as (a) there is the danger of banding, and ( while the image with lower fewer levels will look better in its "raw" state, the image with more levels will generally look better after being passed through noise reduction software, as there's more information to work with.

 

Sandy

Yes, that is clear. However, with a properly exposed ISO 2500 image on the M9 I cannot see banding, nor do I feel the need for noisereduction in post-processing, other than setting it to optimum in the RAW converter.

[sandymc]

Advertisement (gone after registration)

Thanks to Scott, we now know the answer to the new firmware question - no difference; we still have the variable black levels on all images, compressed and uncompressed. Sigh.

 

Sandy

[Jamie Roberts]

Thanks for all of the information. I like to know the options

 

Although it's too soon to say for me (I'm just generally really impressed with the improvements from M8 original to M9 on many levels) I am finding what Jaap says to be true, that even in slightly sub-optimal shots the black point variation is actually producing a better shot overall, at least for practical purposes.

 

Of course, in the dark at high ISOs, I'm very likely shooting something where an aggressive blackpoint is the first thing I'd set in post anyway--so this is saving me a step in essence....

 

I uploaded a late evening shot compressed DNG from the 35 Lux and M9 at ISO 2500 to yousendit for Sandy to look at. There's lots of noise there, to be sure, but C1 makes it a usable file that I can send to printing. I wouldn't even attempt a shot like (well, with this exposure, anyway) on the M8.

 

ETA--Scott beat me to it!

 

[Jamie Roberts]

Thanks to Scott, we now know the answer to the new firmware question - no difference; we still have the variable black levels on all images, compressed and uncompressed. Sigh.

 

Sandy

 

Compressed and *uncompressed?* Does this contradict what you said above, that it won't have an affect on uncompressed DNGs?

[sandymc]

Compressed and *uncompressed?* Does this contradict what you said above, that it won't have an affect on uncompressed DNGs?

 

For uncompressed files, there are 16384 levels available, so losing a few is neither here nor there. Compressed only has 256, so losing a few levels is potentially significant.

 

Regards,

 

Sandy

[Jamie Roberts]

For uncompressed files, there are 16384 levels available, so losing a few is neither here nor there. Compressed only has 256, so losing a few levels is potentially significant.

 

Regards,

 

Sandy

 

Thanks Sandy! I'll bear that in mind for sure.

[jonoslack]

Hi There

I don't have anything significant to add to this thread except:

1. thank you everyone - I now understand why I was finding less noise in jpgs at 2500ISO

2. Jamie - you got an M9 - I missed that and I'm really pleased

[Rolo]

Jamie,

really pleased that your M9 has arrived. Enjoy it, as I know you will.

[Jamie Roberts]

Rolo & Jono--Thanks to both of you! I'm already excited about how great it really is, and I can't wait to put it to good use

 

For those of you in Toronto, I should add that Mike at Eight Elm took really good care of me in getting the M9 too...

[35mmSummicron]

Me too. they run a great little shop

 

[barcoder]

Can you summarize your tips for low noise/high ISO pictures on the M9?

 

I'd like to add this to the M9 wiki.

 

Thanks!

[wlaidlaw]

Sandy,

 

What parameters do you think the M9 then uses to set its black point? Is this done at the white balance stage?

 

Wilson

[sandymc]

Sandy,

 

What parameters do you think the M9 then uses to set its black point? Is this done at the white balance stage?

 

Wilson

 

Wilson,

 

So far as I can tell, it's entirely driven by the ISO setting.

 

Regards,

 

Sandy

[Advertisement]

Hallo,

Look here: Leica M

[t024484]

Black Point is a DC offset level in the Sample values coming from the camera, that should be substracted from all pixel values later on in the digital domain.

In the M8 the black level is compensated, and there is no offset.

In the M9 this DC offset is amplified as part of the pixel value when the ISO setting is changed.

High ISO means a high offset level, low ISO has low offset.

In a compressed DNG, this Black Level offset is fully kept and also compressed.

 

Generating this Black Level offset the way Leica does, makes it possible to find the exact statistical average for the black point, which would not be possible when the signal was chopped at some estimated point. Having Analysed many fully black M9 Images, it is noticeable that the exact black point is gradualy shifting from line to line, from top to bottom. Maybe this has to do with the sensor Temp that is rising during readout.

Anyhow, adjusting the black point with some self produced software to its best possible value, leads to some strange effects in Lightroom, proving more or less that Lightroom is making its own calculations that should not be disturbed.

 

Hans