Is M8 16 or 8 bit?

[Zurenborger]

Advertisement (gone after registration)

there seems to be some confusion about this, with me anyway. According to pre-production data sheets the M8 was planned as a 16bit camera, or am I totally mistaken, I just read that it is an 8bit....

 

could somebody fill me in please ....

[Advertisement]

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

[scott kirkpatrick]

According to pre-production data sheets the M8 was planned as a 16bit camera, or am I totally mistaken, I just read that it is an 8bit....

 

The camera instructions and brochures say that the M8 has "16bit resolution," but in fact the raw files contain 8 bits per pixel, obtained by a nonlinear transformation from the original raw data. So it is actually a "16 bit range," not resolution. And really this sensor is has 12-14 bits of signal above its noise level.

 

scott

[Zurenborger]

scott,

 

thanks for you reply.

 

might you elaborate a bit further ....

 

Plainly speaking I read it three times, and don't understand ;-)

 

thanks

 

Johan

[scott kirkpatrick]

OK, I've said this before, but the thread has cooled off and is buried deep. The marketing literature says, incorrectly, that the M8 has 16 bit resolution. In fact, pixels can range in value from 1 to 2^14, but there are only 256 values used in the compact encoding. The values are chosen well, and you are not likely to see this as a restriction.

 

Here's the set of actions that are taken to turn raw data into a DNG file.

 

For each pixel there is initially a 14 bit value read off the chip. It is multiplied by 4 and then its square root is taken. The square root of a 16 bit number cannot have more than 8 bits, so it fits into one byte in the DNG file. When the DNG file is turned into something like a TIF, it is squared again, and divided by four.

 

The DNG file is an open, publically disclosed format. One of the posters on this forum read some of the M8 DNG files and discovered the lookup tables used to carry out the transformation. Tom Knoll of Adobe confirmed that this is what happens in writing a dng file. So it is not a mystery, but Leica has never provided the promised explanation of why they chose to do this.

 

scott

 

edit: Oh, the noise level part. If you read the imager spec, it says that the pixels can hold a maximum of 60,000 electrons (that is about 2^16), but that the noise level that you expect is about 15-20 electrons, added or subtracted to the pixel by effects that did not depend on the incident light. So the roughly 4 lowest bits in the data are not meaningful, except perhaps when you average many readouts . Readout actually occurs in terms of a voltage, not by actually counting electrons.

[nemeng]

There was a very long thread about this a while ago.

 

A handful of people follow the SK line and say its an 8-bit camera in 16-bit clothing. Others take Leica and Jenoptik at their word and say it's 16-bit. IMO no-one can prove anything unless they have direct access to the RAW data coming off the sensor, before it's touched by firmware and processed into DNG files.

 

I wrote to Jenoptik a couple of weeks ago to ask them to clarify this point, but received no reply.

 

Back when the original thread ran white-hot, the M8 wasn't yet released, so a lot of the argument was necessarily theoretical. But now the M8 is out, maybe there is an independent engineer somewhere who's prepared to strip the camera and apply probes.

 

Don't hold your breath though... :?)

[frc]

Away from the green blobs etc. I find the images even the webposted ones very impressing. Digital Kodachrome I called it before. After the fix of the IR problems I will emediately buy this cam.

Never have I rationaly scrutinized the images a Leica M gives me, they just and only seem very appealing to me. The M8 brings this in to the digital world, question: how relevant is it to know which bit depth it has. Perhaps it is only of no importance to me but an issue to others, just don't know.

File looks good: file looks good.

Naive?

Just don't know,

Problems fixed, happy as a child.

 

 

Greet,

 

 

Fr.

[scott kirkpatrick]

Advertisement (gone after registration)

I repeating myself, but

 

the analysis from a few weeks back was by reading actual M8 files, made by 1.06 firmware, so it was hardly theoretical -- each file consists of 8 bit values, and the lookup table is there as well. As far as how many bits of data there were originally, you could have 32 bits if you wanted, but they represent only 60,000 electrons at maximum, and 15-20 of them are noise. That's from the chip spec. So the actual accuracy that Kodak promises is about 12 bits.

 

I will be most interested to hear what Jenoptik has to say. If you hear from them, please share it.

 

scott

[Srf_Cast]

What’s the bottom line from the Photographer’s perspective?

Can the DNG file expanded in the converter contain any more information than the regular jpg? If it’s 8 bit, than does that mean that the extra information gets invented or guessed in the end?

How many stops of additional dynamic range can the 8 bit DNG vs. jpg contain?

Is this like shooting jpg and than converting in Photoshop to 16 bits? If not, how is it better/worse?

[ho_co]

I'm with those who don't understand this a bit, and I really appreciate the efforts you've all made to clarify the matter again and again.

 

Unfortunately, I took the time (over 2 hours) to look up that hideously long old thread. Read it at your own risk.

 

http://www.leica-camera-user.com/digital-forum/7498-m8-why-10mb-vs-dmr-20mb.html

 

--HC

[leicapfile]

Boy,

 

The more I see threads like this concerning digital, the more I love film.

 

Jerry

[scott kirkpatrick]

What’s the bottom line from the Photographer’s perspective?

Can the DNG file expanded in the converter contain any more information than the regular jpg? If it’s 8 bit, than does that mean that the extra information gets invented or guessed in the end?

How many stops of additional dynamic range can the 8 bit DNG vs. jpg contain?

Is this like shooting jpg and than converting in Photoshop to 16 bits? If not, how is it better/worse?

 

Those are the right questions to ask. While the "information" in the file may be limited by the fact that there are only 8 bits to start with per pixel, those 8 bits are better spread over the full 12-14 bit dynamic range of a scene. There are more levels available for describing shadow detail, for example. When you shoot jpg and try to modify it with a tone curve, posterization can easily result because there is not enough information in the shadows to start with, while there are more levels than are really needed to describe highlights. This is a much better balanced starting point. And of course, the M8 files are quite lovely to start with, and do not seem to need much manipulation in normal scenes.

 

"Guessing extra information" to fill in the spaces between the 256 values that now represent the range from 1 to 4096 -- well if you need to do this, you can add random low order bits to fill in those spaces. Capture One has a control ("banding removal") that does exactly that. The technique is also known as "dithering."

 

As for the gruelling thread in which we first went through the dng decoding, that consisted of three people trying to understand the file content in slow steps by asking each other questions and sharing what we found. For every constructive post, we received three or four howls of "You loonies, Leica would never do that to us!" which had to be answered.

 

scott

[Srf_Cast]

Thank you Scott for the help. It’s still confusing, I’m afraid.

Spreading the 8 bits over 12-14 dynamic range part is especially confusing as I can spread the 8 bit jpg over 16 bit tiff in Photoshop with 1 click of the mouse.

The question still stands. Does Leica DNG offer anything extra in real life over the jpg or is this just a waste of the disk space and time?

[ho_co]

For every constructive post, we received three or four howls of "You loonies, Leica would never do that to us!" which had to be answered.scott

You're right, Scott. But we loonies had to maintain contact with each other while you guys were out dithering!

 

--HC

[scott kirkpatrick]

The question still stands. Does Leica DNG offer anything extra in real life over the jpg or is this just a waste of the disk space and time?

 

Yes, I think it will stand editing much better than will a JPEG and you will get access to the 12-14 bit actual range of the file. In contrast, a JPEG will clip some of the range. But the DNG does not provide the full resolution (microcontrast).

 

scott

[reddawn]

I repeating myself, but

 

the analysis from a few weeks back was by reading actual M8 files, made by 1.06 firmware, so it was hardly theoretical -- each file consists of 8 bit values, and the lookup table is there as well. As far as how many bits of data there were originally, you could have 32 bits if you wanted, but they represent only 60,000 electrons at maximum, and 15-20 of them are noise. That's from the chip spec. So the actual accuracy that Kodak promises is about 12 bits.

 

I will be most interested to hear what Jenoptik has to say. If you hear from them, please share it.

 

scott

 

scott, I'm no electronics engineer, but does this mean the M8 DNG has absolutely no advantage over say, a 5D or 1dsMk2 file when both are converted from RAW to 16 bit TIFFs?

 

The DMR was supposed to have an advantage over its competitors with a native 16bit RAW output.

[scott kirkpatrick]

As far as I can tell, the marketing hype about number of bits is simply hype, and you should look at actual pictures to see which renderings you prefer. The Kodak chips in the DMR and M8 have published specs, so that we know that there are probably 11-12 bits of true information in the 14bit data (packed into the 16bit field of a TIFF) that they generate. You can't print 12 bits, or see it on a screen. Both, as far as I can see, look wonderful when used well. The Canon CMOS chips are proprietary, so we don't know what their output looks like. And Canon uses an AA filter and does noise smoothing at high ISOs. Sounds horrible, but people love what they get with a 5D as well.

 

I think these particular technical specifications are only useful as a starting point in comparing what the cameras will do for you.

 

scott

[adan]

Scott: Nicely handled!

 

"I think these particular technical specifications are only useful as a starting point..."

 

Exactly. How many photographers ever bothered to ask the size of Tri-X grains in microns? You shot it - you shot a little HP3/4/5 for comparison - you tried Microdol or D-76 or Rodinal and decided which look you preferred - and then you just went out and took pictures.

 

There is nothing wrong with trying to increase one's technical knowledge of digital (or film) in a general way - it helps when trying to understand the ins and outs of something like the infrared sensitivity issue.

 

Just don't lose sight of the subject matter for the bits.

[carstenw]

Boy,

 

The more I see threads like this concerning digital, the more I love film.

 

Jerry

Reciprocity tables, anyone? Film is just as complicated, we just learned to deal with it.

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[cbretteville]

Number of bits does not matter, only image quality does. This from a geek that switches bits on and off for a living.

[rick_dykstra]

Boy,

 

The more I see threads like this concerning digital, the more I love film.

 

Jerry

 

Onya Jerry! This last week has been one of the most interesting on the web I can remember. Partly because I haven't got an M8, nor one coming. Yet. hehe.

 

Fascinating stuff. That's Leica for you.

 

Regards, Rick.