Leica M Monochrom review

[hjulenissen]

since the luminance information is only being sampled from the green cells

...With all due respect, all I can say is that if I have to choose between the technical information I get from the scientists and engineers working at the top digital imaging companies in Europe, Japan, and America, or you, somehow I think you'll end up being the loser.

I think this is the wrong way to face the discussion.

There are a number of perfectly understandable reasons why your idea of what some scientists and engineers thinks may not be accurate or the best explanation. They may have worked for some company that have invested in certain technology, they may have been drunk, you may have misread them, or (perish the thought) the scientists could actually be wrong. The best way to figure that out is to present the arguments and references, and/or to do a practical test.


I suggest you ask yourself this question: If a scene contains only monochromatic light (variation) at the far red or blue end of the spectrum (not exciting the "green" channel), does it contain no luminance? Would you expect that a bright blue light should be rendered as dark grey in a "neutral" B&W rendering?
http://en.wikipedia.org/wiki/Luma_(video)#Use_of_luminance

Y = 0.2126 R + 0.7152 G + 0.0722 B

(Only offered as a rough guide as to how the standard red/green/blue channels contribute to luminance - any given camera will typically have somewhat different response)
-h

[BJL]

Although luminance information is not only green, there is a well documented loss of resolution caused by the spatial averaging involved in generating the two missing color values at each pixel. The dominance of the signal from "green" pixels (which in reality measure a wide spectral range across the middle of the visible spectrum) means that at a red or blue pixel, the luminance part of the demosaiced output depends more on the data from nearby green pixels than on the signal from that pixel itself. EDIT: a post by "h" quantifies this: even at a red or blue pixel, green accounts for about 70% of the luminance information, which in turn is what dominates our perception or resolution END EDIT.

Or if you do not trust theory, just look at the resolution measurement for the X3 style pixels of Foveon/Sigma sensors, which roughly match a Bayer CFA with about twice the total pixel count, and so roughly the same count of locations at which green is measured.

On the other hand, those X3 sensors with no AA filters still show some aliasing artifacts, because there is indeed more to aliasing problems than color moiré, even though that is the most visible and famous aliasing problem. There are also some examples of "luminance aliasing" in discussions in this forum.

In defence of Leica, it is probably not in a position to avoid the deep, dark, dirty secret of CCD sensors --- the noise problems in very deep shadows and at high Exposure Index --- because adopting a better sensor technology would have extremely high costs relative to its very low unit sales. In particular, the Kodak sensors for M cameras have a special design with microlenses offset towards the optical axis, needed to deal with the highly off-perpendicular incident light delivered by some short focal length rangefinder lenses that have their rear elements amd exit pupil very close to the focal plane (no SLR lens does this). It could well be that even if a sensor supplier like Sony could do this, none except Kodak and Dalsa have developed suitable microlens technologies, and it is not cost effective to have a CMOS sensor supplier do it. Note that Kodak and Dalsa uses those offset microlenses in some other sensor too, like recent 44x33mm models, so that the cost is shared over more sensors, whereas a CMOS sensor maker would probably have only the Leica M cameras with a use such microlenses.

[Graeme Nattress]

Any decent demosaicing algorithm will derive some luminance information from the red and blue data where available. There are of course two factors that relate to the reduction in resolution from the pixel count to the measured amount, them being the mosaic pattern and subsequent demosaicing and the optical low pass filter.

Even if you look at the actual green pixels you have available (50% of the total), that gives you a linear resolution of 70% without having to interpolate at all. Any reasonably demosaicing will get that up to 80% or thereabouts and more if you remove the OLPF and allow for greater aliasing issues. Chroma moire doesn't come from the "guesswork" per se, but from the under-sampled nature of the red and blue compared to the OLPF which is generally set for the green pixels and their anti-aliasing needs. Chroma moire is not generally an issue and can be rapidly and adequately dealt with by superior demosaicing algorithms, whereas luma aliasing can not.

When designing new sensors, generally the first "slices" don't have the CFA dyes added and are indeed monochrome. This allows the sensor designer to test the pixels without having to concern themselves with the added complexity of demosaicing, and indeed early sensors of a new design will be monochrome. I don't see any added engineering work to get monochrome sensors, although it will impact the image processing pipeline in camera to deal with the monochrome data to make an image, but that processing will now be simpler than with a Bayer CFA and probably achieved by removing stages or using simpler stages (which may need extra development).

All this said, I love the idea of a dedicated monochrome camera, but I don't see it selling in enough numbers to warrant many manufacturers doing this. I had lobbied for a black and white sensor on the RED's (we wouldn't sell enough to make it worthwhile on the re-engineering of the pipeline and compression), but that was mostly just to answer the age-old question of "what is black and white and red all over?"

Graeme

[BJL]

Graeme,
Thanks for your professional input. Your figure of the CFA reducing resolution by about 20%, plus the exta loss due to the AA filter that the MM also avoids, matches up very well with Michael's eye-ball estimate of 26-32MP: just removing that factor of 80% linear resolution would be comparable to increasing linear pixel count by 100/80, and so MP count going from 18 to 18*(100/80)^2 = 28MP.

[Graeme Nattress]

Also, the OLPFs are generally specified a bit weaker in the digital stills world than we do in motion, which is not surprising given that to fix aliasing on one frame is a pain, but to fix it on a movie is deadly.

Graeme

[michael]

Where did the straw man argument that green is only responsible for luminance come from? That's not what I wrote or have ever written. They are predominantly responsible in a Bayer array, but not totally.

I don't mind being pilloried for my mistakes, but not for what other people have said I said.

Real world a Bayer sensor produces about 75% of the resolution of the same sensor without a colour matrix, which means that the red and blue cells are contributing less than the greens to luminance and position information. The M Monochrom simply proves the theory, as anyone who has tested it against an M9 can verify.

Cheech.


Michael

[Alan Smallbone]

Well monochrome sensors have been around for a long time. Leica did not have to do any engineering in "removing" the Bayer array as a lot of sensors on the market are made without them. I have been using monochrome astronomy cameras for a long time, and use color filters in front on a filter wheel to provide color information. Since there is no Bayer array the resolution will be increased. As to sensitivity that will really depend on the sensor being used, many are not as sensitive in certain ranges of the color spectrum and respond differently so how much like film it will be will depend on the sensor design and whether or not the pixels have micro lenses, which are not to be confused with Bayer filters.

All in all it would be a nice camera to have if it were not so outrageously priced. IMO.

Alan

[Bart_van_der_Wolf]

Graeme,
Thanks for your professional input. Your figure of the CFA reducing resolution by about 20%, plus the exta loss due to the AA filter that the MM also avoids, matches up very well with Michael's eye-ball estimate of 26-32MP: just removing that factor of 80% linear resolution would be comparable to increasing linear pixel count by 100/80, and so MP count going from 18 to 18*(100/80)^2 = 28MP.

Hi,

That 80% is only for the Green filtered sensels. When the smaller contribution from the Red and Blue filtered sensels is also utilized, then luminance resolution can easily go up to more than 90% (93.6% as I've demonsrated here).

This is also confirmed when you plot the MTF curves for the R/G/B channels in the same chart, they are virtually identical when luminance variations are present. The only resolution challenge that exists is when luminance is low contrast, but chrominance is saturated Red and/or Blue. Then the demosaicing algorithm has to work a lot harder to get something useful from the undersampled colors.

Cheers,
Bart

[Graeme Nattress]

"The green filtered cells are responsible for luminance information and the red and blue predominantly for colour information, which when combined with the green allows for full colour images to be derived", so although you say "red and blue predominantly for colour information" which does imply they have some impact on luminance resolution, but "The green filtered cells are responsible for luminance information" implies it's only green. That is where the discombobularity is coming from. I think 75% linear resolution is a bit low, and a lot of that is OLPF related rather than Bayer CFA related. What the information from the red and blue do is vastly increase the accuracy of the green interpolation.

Graeme

[Graeme Nattress]

That 80% is only for the Green filtered sensels.

I was including a decent OLPF in that 80% figure. I don't really think we can count excessive aliasing as resolution.

Graeme

[JerseyT]

It would be interesting to compare monochrome images processed from the Nikon D800E to those from the Leica MM.  Seems to me that you should get similar resolution, while gaining the possibility of flexible post-processing filtering, at about 40% of the cost of the Leica.  Yeah, you wouldn't have Leica lenses but Nikon isn't exactly chopped liver, as they say.

[BJL]

Where did the straw man argument that green is only responsible for luminance come from? That's not what I wrote or have ever written. They are predominantly responsible in a Bayer array, but not totally.

Michael, you have got the important point right (like with the 26-32MP estimate) so the problem is just people jumping at some possibly poor phrasing: "since the luminance information is only being sampled from the green cells..." Maybe you should clarify that wording?

[Petrus]

It would be interesting to compare monochrome images processed from the Nikon D800E to those from the Leica MM. 

There is so much Oohing and Aahing about this wonder camera in different forums that I am also extremely curious about a comparison of this new Leica to D800e and also 5D3. Both with unprocessed B&W from those traditional color cameras (D800 has a B&W setting if I remember correctly?) and tweaked conversions with Lightroom or PS. It is true that using the adjustment sliders in LR or PS to the max does not look pretty, but how many people really are going to use the Leica with maximum density red filters for example, as it was claimed that then it really shines? I am not convinced that there is so much resolution advantage in Leica Monochrom that it overcomes the lost possibility of tweaking the color mapping into B&W all digital color cameras give. Quality is not only the resolution, but also the artistic feel and translation into grayscale, which with Leica is fixed.

[theguywitha645d]

The difference between a M9 and an M9M can easily be measure with a test target. Can't get more "real world" than that.

As far as Bayer interpolation making a big impact on resolving power, I don't have that experience. The two 100% images are taken through the same optics with the same camera that make an interpolated image or a real color image by shifting the sensor by a pixel length. You really need to look to see the difference.

The M9M seems to have a different exposure and/or signal processing going on, at least from the samples I have seen. I also don't know anything about the spectral response. But in evaluation B&W images, they can simply appear to have more resolution because of contrast and the elimination of color information. I also shoot monochrome cameras with filter arrays and the monochrome image always appears more detailed even though there is no difference in the resolving power of the system. Judging resolving power by comparing two "real world" images is pretty much a subjective call.

As far as the M9M, it is rather a disappointment. It is funny that the Japanese get lambasted when they don't (seem to) update their technology, but the Germans get let off easily. I definitely get the rangefinder thing--my favorite type of camera. I get the appeal of a monochrome camera. I am just really underwhelmed by the "new" Leica. They could put a little effort into it.

[Aku Ankka]

Aku,

Thank you for your comments. With all due respect, all I can say is that if I have to choose between the technical information I get from the scientists and engineers working at the top digital imaging companies in Europe, Japan, and America, or you, somehow I think you'll end up being the loser.

Also with all due respect, you may well talk to all kinds of experts, but that does not mean that you are an expert in this field (digital imaging, including sensor operation). You are not. I just wish you could understand it and write your articles accordingly. Now they are filled with mistakes and no offense meant, you're quite arrogant in defending your writing by ignoring the criticism and praising your connections. This behavior is not too dissimilar to so called "fanboy" behavior which pollutes discussions.

Let me quote you again:

Engineering a monochrome sensor equipped camera isn't simply a matter of removing the Bayer array. Though based on the M9 sensor, a significant amount of reengineering at the chip level was required

This is absolutely false. If you knew what you are talking about, instead of just repeating what someone at Leica has told you, you'd know that. Of course, I would be delighted if you would tell of one single thing that would need to be reengineered (at the "chip" level - I assume you mean the sensor, but maybe something else).

And something else from your article:

Foveon technology, where the colour filter layers are stacked vertically

Your lack of knowledge is clear here as well - Foveon does not have any color filter layers. Instead it has a triple photodiode construction where modest color separation is achieved due photons of different energies penetrating silicon to different depths with different probabilities.

And when it comes to resolution, BartvanderWolf has demonstrated in this forum that your estimates are quite false. I have also done my share of measurements and the results have been significantly different than your guestimates. I prefer measurements and science anytime over eyeballing.

Anyhow, to me this new Leica is kind of like Pentax K-01 - a minimum risk new product with minum amount of reengineering required to make it. Pentax took a K-5, crippled it and put a new body around it and that's K-01, Leica took M9, used a CFA-less sensor and that's the new camera.

[Aku Ankka]

Or if you do not trust theory, just look at the resolution measurement for the X3 style pixels of Foveon/Sigma sensors, which roughly match a Bayer CFA with about twice the total pixel count

Why should we trust a theory if it confirms your false assumption of Foveon resolving power? No sensor resolves beyond Nyquist, and Bayer CFA equipped sensors resolve over 90% of Nyquist. Even if we use the conservative 90% figure, it leads to Bayer pixel being worth 81% of a Foveon pixel in this regard. Not 50%.

In particular, the Kodak sensors for M cameras have a special design with microlenses offset towards the optical axis

Nothing "special" about that. All manufcturers have off-axix microlenses on mirrorless cameras, and possibly even in the DSLRs (though there the benefits would be significantly less).

elements amd exit pupil very close to the focal plane (no SLR lens does this). It could well be that even if a sensor supplier like Sony could do this, none except Kodak and Dalsa have developed suitable microlens technologies

Wrong. All manufacturers have this capability. Off-axix microlenses are standard technology. And to be even more precice, Kodak and Dalsa are far from being anywhere near the leaders of the pack regarding microlens technology.

Also, they are part of the "toppings", not integral to the sensor itself. Leica could make them themselves (though they don't necessarily have the tools inhouse and the expertise in this special field is certainly lesser than of Sony of other major players in image sensor bussiness).

[Rob C]

From the perspective of someone who's interested but can't afford either type of new(ish)/new M camera (and thus without horse in this race) I can't see the rationale with this camera. Yes, it woud be nice to have something giving 'better' b/w results, but at the expense of the controls now taken for granted with Photoshop, it seems sort of perverse, as does the price, where for something that is giving you less (no colour possibility) you'd really expect, instead, to pay less, a hell of a lot less!

But that's what it means to be a legend: you get star status until your last flop and then you are right off the world's radar.

I think this thing is a mistake, if only because it should have been a lot less expensive than the cooking version of M9.

Rob C

[John R Smith]

I think this thing is a mistake, if only because it should have been a lot less expensive than the cooking version of M9.

The snag is, Rob, the M9 and M Monocrom probably cost almost exactly the same to manufacture. But the Monocrom will sell in far fewer numbers. If Leica cut their margins and sold the Monocrom for less, would they gain enough sales to make a business case? Probably not.

John

[Bryan Conner]

Also with all due respect, you may well talk to all kinds of experts, but that does not mean that you are an expert in this field (digital imaging, including sensor operation). You are not. I just wish you could understand it and write your articles accordingly. Now they are filled with mistakes and no offense meant, you're quite arrogant in defending your writing by ignoring the criticism and praising your connections. This behavior is not too dissimilar to so called "fanboy" behavior which pollutes discussions.

Let me quote you again:
This is absolutely false. If you knew what you are talking about, instead of just repeating what someone at Leica has told you, you'd know that. Of course, I would be delighted if you would tell of one single thing that would need to be reengineered (at the "chip" level - I assume you mean the sensor, but maybe something else).

And something else from your article:Your lack of knowledge is clear here as well - Foveon does not have any color filter layers. Instead it has a triple photodiode construction where modest color separation is achieved due photons of different energies penetrating silicon to different depths with different probabilities.

And when it comes to resolution, BartvanderWolf has demonstrated in this forum that your estimates are quite false. I have also done my share of measurements and the results have been significantly different than your guestimates. I prefer measurements and science anytime over eyeballing.

Anyhow, to me this new Leica is kind of like Pentax K-01 - a minimum risk new product with minum amount of reengineering required to make it. Pentax took a K-5, crippled it and put a new body around it and that's K-01, Leica took M9, used a CFA-less sensor and that's the new camera.

I do not understand why one person would choose to challenge the statements of an easily identifiable person while choosing to hide anonymously behind a fake name.  What do you have to hide?  Or, are you afraid of something?  You would gain a lot of credibility by being brave and come out of hiding.  Be brave young duck!   

[bobtowery]

Is it just me, or are there way more engineers (pseudo-engineers?) around here than photographers?  Does any of this discussion really matter? Would it change how you use the camera? "Excuse me, could you turn a little to the left? Due to the 70% ratio of green luminance and the inherent gain in linear resolution thus rendered, I believe my image will be slightly better. Thank you."

The camera is what it is. If you can afford it and like the way it works, go buy one.

If not, again, does it matter?

There. I feel better. I'm now going to take my Canon for a walk.