Leica M10 raw file (DNG) analysis

[scott kirkpatrick]

Advertisement (gone after registration)

There are (I think) three Exif tags referring to the maximum aperture. First there is the MaximumApertureValue tag with a rational value; this one appears to be correct. Then there is the LensModel tag with a string value; this appears to be correct as well. And then there is the LensSpecification tag with up to four rational values for the shortest focal length, the longest focal length, the maximum aperture for the shortest focal length, and the maximum aperture for the longest focal length. For a prime lens, only two values would be specified, the focal length and the maximum aperture. This is the value that the M10 gets wrong, for whatever reason. I guess it is something trivial.

 

(in response to my question -- "how do you read the files?")

 

Just ExifTool.

 

I took a look at some SL files this morning, and mjh and ExifTool are on the right track.  Here's what the beginning of the main image part of the SL file contained:

 

ExposureTime: 1/125 sec

FNumber: f/4.00

ExposureProgram: Aperture Priority

 

ISOSpeedRatings: 160

 

and a bit lower:

 

ExposureBiasValue: 0.00

MaxApertureValue: f/2.8

...FocalLength: 28.0 mm

 

and much lower down:

 

LensSpecificationExif: 28.0 mm f/3.0

LensMakeExif: "LEICA CAMERA AG"

LensModelExif: "Elmarit-M 1:2.8/28 ASPH."

 

The stuff at the end is hard-coded into the camera firmware, and supplied as the file is written.  The stuff that appears earlier is actually captured by the camera during each exposure.  Even in the SL, the "LensSpecificationExif" is redundant information that gets so little respect that they only got the first digit correct! (in a shot taken with my 28 Summilux, this tag says 28.0 f/1.0).  

 

Apparently in building the M10 firmware table, they have not yet gotten around to filling this entry and some value from whatever that table was used for before has crept in.

 

BTW, the M10 files do not have any entry with the tag "FNumber," so this information has indeed been suppressed.

 

scott

[wlaidlaw]

Does not the external CdS sensor form part of the Thyristor control of the flash system and is used to read the reflected light from the pre-flash. 

 

Wilson

[waterlenz]

Does not the external CdS sensor form part of the Thyristor control of the flash system and is used to read the reflected light from the pre-flash. 

 

Wilson

I would hope not as this would mean that strobe exposure control was not actually TTL.

[Erik Gunst Lund]

Re the small light sensor on the front of the M8, M9, M and M10.

 

The principle was patented back in 2006 by Leica

Although it wasn't a US patent until Oct. 15. 2009

 

DIAPHRAGM VALUE RECOGNITION US 2009/0256954

Scope of the Patent:

A method for determining the current camera aperture (working aperture) of lenses on digital cameras having a viewfinder eyepiece and internal exposure measurement through the lens is wherein a further external exposure measurement is carried out past the lens and the value of the working aperture is determined from a reference table stored in the digital camera from the difference Delta BV between the two measured exposure values BV(int) and BV(ext).

 

 

I have found it to be very accurate on M8 and M9, much better than I initially thought possible

[jaapv]

I would hope not as this would mean that strobe exposure control was not actually TTL.

No, it does not interfere with TTL accuracy. It determines the intensity of the pre-flash, as the camera needs to get the light level within the measuring range. You can notice that sometimes, when the camera gets the pre-flash intensity widely wrong, it will do a second pre-flash.

[tom0511]

I agree it would be good to have it back. Its not accurate but at least gives an idea.

[sandymc]

Advertisement (gone after registration)

I took a look at some SL files this morning, and mjh and ExifTool are on the right track.  

 

 

Yes, as I read the specification, the values in the LensSpecification tag are static for a particular lens. Most probably Leica or whoever they subcontracted the software to just didn't fill out the table correctly. The "light sensor" value should not have any effect.

 

Sandy

[sandymc]

Great service, thank you, Sandy.

 

You note the color matrix in the RAW file suggests sensor is different.

 

Would not the same sort of differences be seen with a different choice of IR cut in the coverglass, since the IR cutoff curves vary greatly between say a schott S8612 and BG55 in the visible spectrum?

 

IE the sensor is the same but IR cut has been changed.

 

It would be interesting on this same note to compare M9 files before and after the current sensor upgrade which involves such a change.

 

Could you elaborate a bit on the color matrix, and it's relation to sensor design, or suggest a link?

 

Thanks so much

 

 

The primary driver of the matrix is the characteristics of the dyes used in the Bayer array. In addition, the effects of IR filters, the actual sensitivity of the "naked" sensor, etc all come into it. Why there is a matrix is because the spectral response of the dyes overlap. In simple terms, a bit of "red" light gets into the "green" pixel, a bit of green light gets into the red pixel, etc. The matrix subtracts that out. If you look at a matrix, the terms on the diagonal are usually positive, while the other terms are smaller and negative. The smaller negative terms are subtracting out the overlaps in spectral sensitivity. As such, there is no "perfect" matrix, just the best compromise you can find.

 

Sandy

[waterlenz]

No, it does not interfere with TTL accuracy. It determines the intensity of the pre-flash, as the camera needs to get the light level within the measuring range. You can notice that sometimes, when the camera gets the pre-flash intensity widely wrong, it will do a second pre-flash.

Thanks I did not realize that pre-flash intensity can vary according to ambient light level. I have not detected this with N and C digital TTL but I do most of my flash photography manually. However, I took the post I was replying to mean that the external sensor was used to determine flash duration, i.e., no data from the sensor in the focal plane. The poster said "used to read the reflected light from the pre-flash."

[dkCambridgeshire]

Could the Viso 'blackout' anomaly be remedied via a firmware fix?

 

dunk  

[uhoh7]

The primary driver of the matrix is the characteristics of the dyes used in the Bayer array. In addition, the effects of IR filters, the actual sensitivity of the "naked" sensor, etc all come into it. Why there is a matrix is because the spectral response of the dyes overlap. In simple terms, a bit of "red" light gets into the "green" pixel, a bit of green light gets into the red pixel, etc. The matrix subtracts that out. If you look at a matrix, the terms on the diagonal are usually positive, while the other terms are smaller and negative. The smaller negative terms are subtracting out the overlaps in spectral sensitivity. As such, there is no "perfect" matrix, just the best compromise you can find.

 

Sandy

Thanks so much for taking the time to explain this, Sandy

 

So the IR cut is unlikely to account for such differences, but you might have the same "naked sensor" but a new CFA and see changes in the color matrix as we are?

 

I note some CMOS providers now tout "in house" CFA:

http://www.towerjazz.com/cmos-image-sensor.html

 

I'm wondering what are your thoughts about how the options of CMOS sensors are evolving and what clues the DNGs may give us as to Leica's choice of technology.

 

Puts initially was of the belief this was the SL sensor with better noise algorithms. In and update he said the "basic sensor" meaning naked sensor may still be essentially the SL sensor yet with some new choices in common technology, like I suppose the CFA.

 

SL sensor is supposedly from a new manufacturer, not CMOSIS. Which implies the SL color matrix is very different from M240 DNG also, did you see this?

 

TY so much

[sandymc]

Thanks so much for taking the time to explain this, Sandy

 

So the IR cut is unlikely to account for such differences, but you might have the same "naked sensor" but a new CFA and see changes in the color matrix as we are?

 

I note some CMOS providers now tout "in house" CFA:

http://www.towerjazz.com/cmos-image-sensor.html

 

I'm wondering what are your thoughts about how the options of CMOS sensors are evolving and what clues the DNGs may give us as to Leica's choice of technology.

 

Puts initially was of the belief this was the SL sensor with better noise algorithms. In and update he said the "basic sensor" meaning naked sensor may still be essentially the SL sensor yet with some new choices in common technology, like I suppose the CFA.

 

SL sensor is supposedly from a new manufacturer, not CMOSIS. Which implies the SL color matrix is very different from M240 DNG also, did you see this?

 

TY so much

 

 

Yes, you might have the same "naked sensor" with different dyes, and get a different color matrix. 

 

This is the first color matrix from an SL:

ColorMatrix1:      
      1.2587  -0.5232  -0.1496
      -0.3610   1.0841   0.0277
      -0.0911   0.1674   0.2070

This is from a M240:

ColorMatrix1:
      0.7571  -0.2249  -0.0223
     -0.4171   1.3145   0.1062
     -0.0357   0.1601   0.7591

And finally from an M10

ColorMatrix1:
      0.8950  -0.3206   0.0493
     -0.5830   1.6099  -0.0022
     -0.0837   0.2769   0.7937

So yes, the SL matrix is quite different to the M240 matrix

 

Sandy

[grillo]

I know it's meaningless, but just for fun, in terms of euclidean distance the M10 Matrix is much closer to the M240 matrix than to the SL matrix. In fact, the M240 is closr to the SL than the M10:

 

dist(M10, M240) = 0.42

dist(M10, SL) = 0.95

dist (SL, M240) = 0.85

[ramarren]

then ignore the info.

Your statement made no sense in the context of what I wrote that you quoted. I asked a question.

[bencoyote]

 

For those interested in the gory technical details, my usual "new Leica camera" analysis is up. Although in this case there's quite a lot that's (as yet) unknown.

 

Leica M10 raw file (DNG) analysis http://chromasoft.blogspot.com/2017/01/leica-m10-raw-file-dng-analysis.html

Sandy

 

Thanks for looking into that. I was kind of curious about that myself.

 

I'm kind of surprised that it is only 14-bits per pixel.

To me that suggests that there is no additional dynamic range in the M10 over the M type 240 because the maximum theoretical range would be 14 stops and the M already approaches that according to DXO Mark (13.3).

 

Given the association between ISO and DR and the fact that they were able to boost the ISO range almost 3 stops, I would have thought that the A-to-D converters along with their associated amplifiers were suitably improved that they would have changed to 16-bits per pixel to get more color depth at the lower end of the ISO range. Do you have any idea why this may not have been the case?

[uhoh7]

Yes, you might have the same "naked sensor" with different dyes, and get a different color matrix. 

 

This is the first color matrix from an SL:

ColorMatrix1:      
      1.2587  -0.5232  -0.1496
      -0.3610   1.0841   0.0277
      -0.0911   0.1674   0.2070

This is from a M240:

 

ColorMatrix1:
      0.7571  -0.2249  -0.0223
     -0.4171   1.3145   0.1062
     -0.0357   0.1601   0.7591

And finally from an M10

 

ColorMatrix1:
      0.8950  -0.3206   0.0493
     -0.5830   1.6099  -0.0022
     -0.0837   0.2769   0.7937

So yes, the SL matrix is quite different to the M240 matrix

 

Sandy

 

Thank You so much, sir

[sandymc]

Thanks for looking into that. I was kind of curious about that myself.

 

I'm kind of surprised that it is only 14-bits per pixel.

To me that suggests that there is no additional dynamic range in the M10 over the M type 240 because the maximum theoretical range would be 14 stops and the M already approaches that according to DXO Mark (13.3).

 

Given the association between ISO and DR and the fact that they were able to boost the ISO range almost 3 stops, I would have thought that the A-to-D converters along with their associated amplifiers were suitably improved that they would have changed to 16-bits per pixel to get more color depth at the lower end of the ISO range. Do you have any idea why this may not have been the case?

 

Well, I don't much trust DxO ratings, I'm afraid.

 

The way that Leica would have looked at bit depth, in my view, would be to look at the noise floor of the sensor, and take a decision whether the additional 2 bits of A/D resolution actually bought them anything, or would just be digitizing noise. That of course assumes that Leica had any real say in the design of the A/D. Typically A/Ds on CMOS sensor are tightly integrated with the sensor design. For Leica to customize the top levels (Bayer array, IR filter, micro lenses, etc) of a sensor is one thing, to mess with the "core sensor" is another thing.

 

Sandy

[Erik Gunst Lund]

To capture in 16 Bit you would need to buy a Phase One XF 100MP camera

[IkarusJohn]

To capture in 16 Bit you would need to buy a Phase One XF 100MP camera

Oh, fighting talk! I love it when some one talks dirty!

[Advertisement]

Hallo,

Look here: Leica M

[Erik Gunst Lund]

It's Danish