Hi Christopher,
The reason is the MF back manufacturers they are lying.
Cheers
Simon
Can you explain?
Henrik
Thanks,
but what then contributes to making MF files so much more robust to post processing...
Thanks,
but what then contributes to making MF files so much more robust to post processing. Is it the calibration related to iso so that if I underexpose my DSLR I get the same effect, and/or is it the size of the sensor?
Christopher
but what then contributes to making MF files so much more robust to post processing.
Suffice it to say that medium format cameras don't have low cost, high ISO, fast shooting speeds, or a huge number of features, and they are more difficult to learn and to use; so if they don't deliver fantastic image quality and a good user experience then they won't be purchased by anyone. As a result the engineering, marketing, and resources of medium format companies goes very heavily into making the image capture the best quality images (even if it means sacrificing a convenience or non-quality-related feature).
Ii recently acquired a leica S2. On a recent trip to the Eastern Sierras (California), I was fortunate enough
to witness an incredible sunset, and in particular the "after glow" over the mountains. The sky was painted
with shades of red, yellow, pink, orange, more vivid and varied than I have ever seen. Along with the S2,
I had a nikon D700. I set both up at the same time, and took repeated images. The D700 could not capture
the colors to the extent that the Leica S2 did. The range of colors, the tonal gradations, the gradual shifts
from one color to the next, was clearly superior on the S2 vs. the D700. It was visible on my calibrated
monitor, and even more so on a print.
Indeed, throughout this trip, the "micro contrast" and tonal range was clearly superior on the MF S2 than
on the D700. Whether it was rocks, desert sand dunes, salt crystals on the salt flats (Death Valley), there
was a clear distinction. I do not mean merely in terms of resolution, but in the fine tonal contrast that
lends "texture" and a 3-D appearance to the objects in the photo.
Obviously, this is NOT a scientific study, but it was a side-by-side comparison. I cannot explain why there
is such a distinction, whether it is CMOS vs. CCD, 14 vs 16 bit, or the algorithms used to interpret the
collected photons. It is just an observation. When I show the images to colleagues they too can identify
the S2 vs. the D700 images.
I think the three big factors in your example are (1) The superb S2 lenses + (2) the lack of an AA filter together provide the remarkable microcontrast and tonal texture; and (3) Kodak's Bayer-CFA filter bandpasses provide the remarkable colour range/gradations. It has little or nothing to do with CMOS vs. CCD or 14 vs 16 bit, especially since the subject matter you describe is at the mid and high end of the histogram.
Hi,
A sensor cell can hold say 30000 - 60000 electrons. Larger pixels hold more electrons. So upper limit is 60000, which correspond to 16 bits. Now, this signal needs to be read out and converted to binary digits. The readout has some noise. Some CMOS sensors do it on chip, and they can have as low noise as 1-2 electrons, CCD-s have much higher readout noise. more like 15 electrons.
So the calculation is:
MF maximum signal is perhaps 60000 electrons, readout noise is 15 electrons. SNR (Signal Noise ration) is 60000/15 -> 4000 -> 12 bits.
DSLR (Nikon D3X)
Maximum signal 30000 electrons readout noise 2 electrons, SNR = 30000 / 2 -> 15000 which is about 14 bits.
So in this case the Nikon would actually be 14 bit while the MF back would be 12 bits. The figures are approximately in the ballpark.
Would MF backs make good use of sixten bits they would also offer excellent high ISO capability
Best regards
Erik
Exactly my point.
I just performed a small test. I took the same image with a Canon 5D II and a Leaf Aptus II 12 back (both at base ISO). I matched the histograms as close as I could, and deliberately underexposed. I also used an object with dark detail.
Then I imported both RAW files into C1 and pushed them both 2 stops. It's not just the noise performance which is better with the Leaf, but the colour is far superior:
Erik,
That's about right for the MFD system, but even though I know you're using ballpark figures, you're still somewhat off on the D3X for two reasons:
1) It's not as good as say a Canon 1d MkIV in terms of minimum readnoise; the D3X is closer to 4 electrons than 2. That in itself is a drop of 1 bit of DR.
2) Your D3X DR calculation assumes that the minimum readnoise is attainable AT THE SAME TIME as the full pixel capacity (well depth) is also attainable. But because of A/D converter noise, the lowest readnoise in DSLRS is usually only reached at around ISO 800 or 1600, at which setting the maximum signal is reduced by around 4x - 16x, depending on the base ISO.
So the calculation you did correctly gives the sensor's inherent DR, but fails to take the rest of the real-world camera system into account (the A/D contribution to noise). You are in good company: I've seen people quote from Roger Clark's DR tables and plots (http://clarkvision.com/articles/digital.sensor.performance.summary/index.html) while missing the crucial nuance that these are sensor not camera values, and I know I've done it wrong myself in the past :-[. It's rather like saying that "because I can run a short 100m race at 5m/sec, and because I can also run a marathon, then I can run a marathon at 5m/sec" ["because I can readout a truncated max signal at very low readnoise, and because I can also store a max signal up to the full well depth, then I can readout the full well depth at very low readnoise"]. It's combining performance specifications taken under different, mutually exclusive circumstances. I guess it's an easy mistake to make, because people like you and I grew up on the strict engineering definition of DR=FWC/RN, which is fine for CCDs as the readnoise rarely changes with ISO, and ISO settings are usually only "flags" which do not actually decrease the maximum signal (...and scientific CCDs have no concept of ISO in the first place!). But that definition needs to be adjusted for CMOS sensors with real ISO, max signal and readnoise variations.
Just now Doug reminded us again that "it's the system that matters", and he's absolutely right; but I think that he is underestimating the importance of the A/D converter as one of the kingpins of performance.
While Nikon did make a breakthrough with the D3X, managing to greatly reduce A/D noise so that it was much less of a limiting factor at lower ISOs, there still is a modest trend with ISO (D3X on Sensorgen (http://sensorgen.info/NikonD3X.html)). More recent cameras from Sony and Pentax are also greatly diminishing the trend of readnoise with ISO, by beating down the A/D component of readnoise. This is how the Pentax K-5 took everyone by surprise with its ~14 bits DR ( K-5 on Sensorgen (http://sensorgen.info/PentaxK-5.html), and likewise the Nikon D7000.
Ray
I am not that surprised, the DR of the 5DII, and therefore its shadow noise, is notoriously not that good. Even the 18MP 7D does nearly as well although the pixels are much smaller:
It's all I had at hand. As for the DR, it is above average in DR performance (according to DxO), and certainly one of the most popular cameras in use by professionals, so it's a pretty good indicator of why 35mm DSLRs have a particular reputation.
A larger sensor will have better MTF for fine detailsWould you care to elaborate? Do you mean that the pixel count is higher or that MF lenses tends to have higher MTF cutoff (lp/ph)?
Would you care to elaborate? Do you mean that the pixel count is higher or that MF lenses tends to have higher MTF cutoff (lp/ph)?
-h
I just performed a small test. I took the same image with a Canon 5D II and a Leaf Aptus II 12 back (both at base ISO). I matched the histograms as close as I could, and deliberately underexposed. I also used an object with dark detail.
Then I imported both RAW files into C1 and pushed them both 2 stops. It's not just the noise performance which is better with the Leaf, but the colour is far superior:
(you will need to use 'view image' to see it at 100% or click on http://moskvamodels.com/images/canon_v_leaf.jpg )
(http://moskvamodels.com/images/canon_v_leaf.jpg)
Thanks for the test.
I am not that surprised, the DR of the 5DII, and therefore its shadow noise, is notoriously not that good. Even the 18MP 7D does nearly as well although the pixels are much smaller:
http://www.dxomark.com/index.php/Cameras/Compare-Camera-Sensors/Compare-cameras-side-by-side/%28appareil1%29/619|0/%28brand%29/Canon/%28appareil2%29/485|0/%28brand2%29/Nikon/%28appareil3%29/483|0/%28brand3%29/Canon
Cheers,
Bernard
I also made a point that I would expect cameras with > 14bit DR to excel at high ISO. I was always somewhat confused by the statement that MFDBs have large DR but don't perform well at high ISO, you may perhaps spread some light on the issue?
Hi,
Magnification will be higher for an MF system. So if an MF system uses an 80 mm lens and a DSLR a 50 mm lens to achieve the same field of view the image on the sensor will be 1.6 times larger, giving significantly higher MTF.
Best regards
Erik
Hi Erik,
No problem. I think the best way to explain this is to build on my last post above. Let's take the 5DII as our "typical CMOS DSLR" in noise terms. It may have only a modest DR~11 stops at both ISO 100 and ISO 800, but they aren't the same 11 stops! If you map them to the eV range of the scene, at ISO 800 the 11 stops are shifted by +3 eV to lower light levels. This shifting continues by about another 0.5 eV if you raise the ISO further. That's a total scene range of 14.5 stops if you combine ISO 100 and ISO 1600 (and keep the same shutter speed). This explains why the low-light, high ISO performance is so good....it's not the absolute DR at any one ISO that matters; it's the faintest eV that the DR can touch.
Contrast that with a "typical CCD MFDB" like Graham's Leaf, which does indeed have a large DR at base ISO: 12 stops according to Leaf; 1 stop better than the Canon. Changing the ISO setting does not shift the Leaf's DR to lower eVs, since the readnoise is already pretty much at its "floor" level at base ISO. So it has a (14.5 - 12 =) 2.5 eV/stops disadvantage with respect to the Canon, in the "high ISO, low light" regime.
Ray
Easy to explain what's going on here. The key thing is that Graham used "both at base ISO". Now, the 5DII at ISO 100 is one of the noisiest cameras you'll ever come across. Sensorgen finds it has a whopping 27.8 electrons of readout noise at ISO 100 - to put that in perspective, it's noisier than even "prehistoric" 35mm full-frame CCDs like the Dalsa ones in the good old Leaf Volare/Cantare of the late '90s!
Graham's Aptus II 12 has probably at best around 13 electrons readout noise - I say probably, because Dalsa never released a datasheet for their 80MP sensor. The nearest actual comparison I can find is the P65+, which has 17.6 electrons readnoise in Sensorgen. So if it's in this range, it's twice as good as the 5DII at base ISO. This is clear from the images too.
So here's the first thing: Graham was really comparing readout noise at base ISO, not DR.
Here's the second thing. Graham underexposed and pushed the files by 2 stops - effectively simulating ISO 400. With the Leaf, actually shooting at ISO 400 would yield the same S/N, since MF CCD readout noise doesn't change with ISO gain. But with the Canon, actually shooting at ISO 400 would vastly improve the results, since the readnoise drops to 8.6 electrons. Now it's well ahead of the Leaf! At ISO 800 and above, the Canon really pulls away further from the Leaf. And all the while, the Canon maintains a DR of 11 stops - both at ISO 100 and at ISO 800. How does it do this? Because the max signal collected happens to fall in direct proportion to the improvement in readnoise (then the readnoise improvement starts to level off). The Leaf, on the other hand, loses a stop of DR with every doubling of ISO; it's lost 3 stops of DR at ISO 800!
Conclusions?
1) What we all knew already: MFD systems are wonderful at base ISO; and fall increasingly behind in both noise and DR at any other ISO [unless pixel binning is employed].
2) People who have a 5DII alongside an MFD system primarily use the 5DII for its wonderful higher ISO performance (well I do anyway!)...and, well, movies.
Ray
Here's the second thing. Graham underexposed and pushed the files by 2 stops - effectively simulating ISO 400. With the Leaf, actually shooting at ISO 400 would yield the same S/N, since MF CCD readout noise doesn't change with ISO gain. But with the Canon, actually shooting at ISO 400 would vastly improve the results, since the readnoise drops to 8.6 electrons. Now it's well ahead of the Leaf!
In the shots of the ruins that I posted about earlier, one could raise the shadows by either changing exposure values, using the shadow tool, or the curves tool. In all cases when the shadows were lifted on the D3X shot they came up fairly clean - meaning if noise was your determinant for DR, you'd think this camera set up was great since it was clean -but there was just mostly black in the window areas that got lighter as they were pulled up. The Aptus files, on the other hand, contained some detail that came out of the blacks when you lifted the shadows. Clearly the Leaf backs captured more range. Overall I was impressed by the D3X files because I think they are better than my 5D2 files (but I have no direct comparison). I suppose the extra pixels give the Leaf the advantage but I'm not sure how much that really comes into play because it wasn't like the extra pixels just reduced the noise because in this case one had detail in the shadows and the other just didn't record that range at all.
Update: I think Ray answered this already in another posting. I don't remove my posting as I think that the figures are quite nice.
Hi Ray,
Thanks for explaining. When I look at DR on DxO I can see that Canons and cameras having Sony's new CMOS sensors behave very differently regarding DR. I enclose three figures for Nikon D3X, Nikon D3S and Canon 5DII.
In my view the Nikon D3X is very different from the other two. I got the impression that Sony Alpha 77, Pentax K5 and other new Sony based cameras are pretty similar to the D3X curve in shape.
The impression I have is that the readout noise on Canon is poor and is helped by pre amps. So DR is less than on say Nikon D3X at low ISO, but by increasing pre amplification it can be maintained at the same, relatively low, level up to say 800 ISO.
The Nikon D3X still performs well at decently high ISOs according to what I have seen and read.
Best regards
Erik
Hi Eric,
Your comment - "but there was just mostly black in the window areas that got lighter as they were pulled up. The Aptus files, on the other hand, contained some detail that came out of the blacks when you lifted the shadows....and the other just didn't record that range at all " - sounds awfully like the Nikon had clipped up to half of those pixels to zero intensity. Nikon does apply an overly harsh zero offset correction to its RAW files, which does not endear their firmware to us.
Ray
Ray,
That could be because the blacks when lifted were very clean. I don't own a D3X and am mostly unfamiliar with it, so thanks for pointing that out. I also note that I used C1 for my comparison so its possible that other RAW converters would show a different result.
Been following this thread quietly and right now all I can say is that I'm glad that photography is made out of photos and not out of graphs...If 16 bits matter, then one would hope that it would be possible to make relevant side-by-sides showing its benefits, and/or to understand why. Graphs and formulas have been used to teach engineering and (one would hope) to build Leaf & Mamiya cameras, why is it problematic to use graphs to discuss merits of cameras?
This is not all just about shadow noise or shadow detail...a good high bit-depth file will also provide finer, smoother gradations across the range especially in bright tones where a lower bit-depth file will tend to either block them or show bands/ steps.
Simply believing sales people or random internet people at face value that may have have done unfair side-by-sides under god knows what conditions is not my thing.
-h
But as I always say, don't believe internet blurb...do your own tests and draw your own conclusions!
..Graphs tends to counteract that human flaw..
And graphs can be misleading when used out of context or when used to tell an incomplete story.
Graham -
If you have a chance, could you shoot the same scene at ISO 400, to see what shows up?
You can't blame the data. The problem is in the interpreter, the human interpreter.
Absolutely true. In my day job I'm writing engineering software, I work with technical issues daily, and I see every day how easy it is to misuse or misinterpret data if the Big Picture isn't considered. This forum has some of the most obsessive number-worshiping I've seen anywhere but lacks the Big Picture vision: "do you like the photos?"
Here is my partial list of the Image Quality Chain:
Lens Hood / Flare > Lens coating > lens > aperture/shutter > body's internal blackness > IR filter > microlenses > AA filter (or lack thereof) > sensor size > sensor pixel type > readout speed > sensor-to-AD-convertor path, A/D convertor (both bit depth and quality) > heat sinking / cooling > raw file compression > black calibration > in camera raw data manipulation > characteristic curve > ICC profile > demosaic algorithm > deconvolution algorithm > noise reduction type > up-res or down-res algorithm > sharpening
Any one of the above can influence the final image. It is a system, and no one component is as important as the overall system.
but I'm pretty sure that Canon applies noise reduction in camera before the raw file is saved.
Quite a good list indeed. A/D converters issues, read noise and well capacity are still worth discussing from time to time imho as they allow debunking of certain advertising claims.
The supposed de-bunking is equally misleading if it doesn't consider the entire imaging chain.
Not necessarily. If it can be shown that any point in the imaging chain makes it physically impossible for the final output to be as claimed, then there is no need to consider the rest of the chain.
Ray
Not necessarily. If it can be shown that any point in the imaging chain makes it physically impossible for the final output to be as claimed, then there is no need to consider the rest of the chain.
Actually, bees can fly. This piece of urban legend come from people applying data and assumptions incorrectly. There is no scientific evidence to suggest the bee is an aero-impossibility.
What do you mean by "less than ideal"?
Absolutely true. In my day job I'm writing engineering software, I work with technical issues daily, and I see every day how easy it is to misuse or misinterpret data if the Big Picture isn't considered. This forum has some of the most obsessive number-worshiping I've seen anywhere but lacks the Big Picture vision: "do you like the photos?"I disagree. I find the technical discussions on this site interesting and sober wrgt real-life im
And bees can't fly. I know this doesn't satisfy the left brain but assumptions often predispose the outcome and IMHO the best way to compare imaging systems is to use them as they'd typically be used and compare the final results.
What I know is that clients, gallery owners, and random visitors to my website make a point of commenting on the clarity, color richness and gradation, and detail of my photos, both in print and on the web, compared with their own and with other gallery prints. I can assure you that my meager processing skills and very basic software are not an advantage. My camera uses a 16-bit ADC. Perhaps under ideal conditions that's irrelevant but under less-than-ideal conditions when I have to push the files' limits it falls apart much less often than files from 12- and 14-bit cameras. Would your left brain like to explain this?
Ii recently acquired a leica S2. On a recent trip to the Eastern Sierras (California), I was fortunate enoughYes, but the same comparison between the S2 and the D3x would come up more even-handed. The D3x with the 24.5MP Sony Exmor on board has that kind of color gradation and polish, and a stunningly noise-free capture. I'd expect the newest Sony 35mm full-frame sensors to be very good indeed.
to witness an incredible sunset, and in particular the "after glow" over the mountains. The sky was painted
with shades of red, yellow, pink, orange, more vivid and varied than I have ever seen. Along with the S2,
I had a nikon D700. I set both up at the same time, and took repeated images. The D700 could not capture
the colors to the extent that the Leica S2 did. The range of colors, the tonal gradations, the gradual shifts
from one color to the next, was clearly superior on the S2 vs. the D700. It was visible on my calibrated
monitor, and even more so on a print.
Indeed, throughout this trip, the "micro contrast" and tonal range was clearly superior on the MF S2 than
on the D700. Whether it was rocks, desert sand dunes, salt crystals on the salt flats (Death Valley), there
was a clear distinction. I do not mean merely in terms of resolution, but in the fine tonal contrast that
lends "texture" and a 3-D appearance to the objects in the photo.
Obviously, this is NOT a scientific study, but it was a side-by-side comparison. I cannot explain why there
is such a distinction, whether it is CMOS vs. CCD, 14 vs 16 bit, or the algorithms used to interpret the
collected photons. It is just an observation. When I show the images to colleagues they too can identify
the S2 vs. the D700 images.
And bees can't fly. I know this doesn't satisfy the left brain but assumptions often predispose the outcome and IMHO the best way to compare imaging systems is to use them as they'd typically be used and compare the final results.There are no 16-bit cameras made today. Nor 15-bit cameras.
What I know is that clients, gallery owners, and random visitors to my website make a point of commenting on the clarity, color richness and gradation, and detail of my photos, both in print and on the web, compared with their own and with other gallery prints. I can assure you that my meager processing skills and very basic software are not an advantage. My camera uses a 16-bit ADC. Perhaps under ideal conditions that's irrelevant but under less-than-ideal conditions when I have to push the files' limits it falls apart much less often than files from 12- and 14-bit cameras. Would your left brain like to explain this?
Ii recently acquired a leica S2. On a recent trip to the Eastern Sierras (California), I was fortunate enough
to witness an incredible sunset, and in particular the "after glow" over the mountains. The sky was painted
with shades of red, yellow, pink, orange, more vivid and varied than I have ever seen. Along with the S2,
I had a nikon D700. I set both up at the same time, and took repeated images. The D700 could not capture
the colors to the extent that the Leica S2 did. The range of colors, the tonal gradations, the gradual shifts
from one color to the next, was clearly superior on the S2 vs. the D700. It was visible on my calibrated
monitor, and even more so on a print.
Indeed, throughout this trip, the "micro contrast" and tonal range was clearly superior on the MF S2 than
on the D700. Whether it was rocks, desert sand dunes, salt crystals on the salt flats (Death Valley), there
was a clear distinction. I do not mean merely in terms of resolution, but in the fine tonal contrast that
lends "texture" and a 3-D appearance to the objects in the photo.
Obviously, this is NOT a scientific study, but it was a side-by-side comparison. I cannot explain why there
is such a distinction, whether it is CMOS vs. CCD, 14 vs 16 bit, or the algorithms used to interpret the
collected photons. It is just an observation. When I show the images to colleagues they too can identify
the S2 vs. the D700 images.
There are a lot of terms describing image quality that are less than well defined. Unfortunately it's not always clear what is meant with the wording. For instance you use the term "microcontrast". I haved been told that it means MTF at high frequency.
Unfortunately, it is not always feasible for ordinary users to determine stuff such as MTF, "microcontrast", etc. Special test charts, methodology and software have to be used. And, still what about real images that you have acquired? I.e., images of landscapes, cats, oranges, etc., and not some test charts in controlled setting. To complicate matters further, arguing what is lens sharpness, pixel pitch, FOV, and what not. At the end of the day it is the image we are after and it will be useful to have notions of image quality purely based upon on pixel data and detached from sensor pitch, lens, aperture, image display size, etc.
Realizing the vagueness and difficulty associated with this paradigm I developed a measure of image detail, JIDM, for which a user has to just run it through Photoshop (or ImageJ) and it gives you a number in [0-1] range, where higher means more detail. As an example see below:(http://djjoofa.com/data/images/jidm.jpg)
The good thing is that one can select an area of an image using Photoshop marquee tool and it will only do the detail measure analysis in that area. You can download it freely from my website. At this stage it is only Photoshop CS3, Mac OS 10.6. It is not perfect, and there is room for improvement, but it helps me in some of my analyses, and may be you find it useful, or have suggestions for improvements.
And, BTW, while you are there, you might like to get hold of Mac versions of FFT/IFFT plugins that are quite helpful in certain situations.
Sincerely,
Joofa
Unfortunately, it is not always feasible for ordinary users to determine stuff such as MTF, "microcontrast", etc.What is an "ordinary" user? Surely, most camera-owning people would call it extra-ordinary to spend $1000 or $100000 on camera gear, and to participate in a discussion on luminous-landscape.com about the real-world advantages of using 16-bit analog-to-digital converters in cameras having a larger image sensor than 24 x 35 mm?
Special test charts, methodology and software have to be used. And, still what about real images that you have acquired? I.e., images of landscapes, cats, oranges, etc., and not some test charts in controlled setting. To complicate matters further, arguing what is lens sharpness, pixel pitch, FOV, and what not.It is always difficult to relate "lab tests" to real-world usage. Is your new car really able to pull 0.36 liters of gas per 10km of your usage pattern? Is your Kenwood really able to output 1200Watts of dough-massaging (or is most of it going to heat and sound?), and what is the relevance for making bread? One reason why people and engineers still use these "synthetic" measurements is that they are/should be universal, repeatable and at least correlated with significant user patterns. I may not drive like whatever EU/US pattern is used to measure car fuel effciency. You may not either. But perhaps the measure is still sufficient robust for me or you to aid choosing a car without making too large errors?
What is an "ordinary" user?
For those that spend the time and resources, and have the expectations that make MTF relevant, I dont see why it is such an obstacle.
What is the MTF of this image?How many litres are there in one meter? I think that we have had this discussion before. I am able to use abstract measurements in my job and private life. A lot of other people are as well. If you cannot (or if you make up a fictious character for arguments sake), then I dont know how to help out, really.
How many litres are there in one meter? I think that we have had this discussion before. I am able to use abstract measurements in my job and private life. A lot of other people are as well. If you cannot (or if you make up a fictious character for arguments sake), then I dont know how to help out, really.
-h
JIDM is a measure of detail in usual photographic, textured images, especially when the traditional MTF notion can't be directly applied.And for that I am sure that it is a fine tool. My objection was that you seemed to think that MTF was too difficult a concept for photographers to understand or use. I think that most "serious" photographers are in fact able to understand complicated technical concepts if they think that it will help them do their thing.
Sincerely,
Joofa
Here is my partial list of the Image Quality Chain:
Lens Hood / Flare > Lens coating > lens > aperture/shutter > body's internal blackness > IR filter > microlenses > AA filter (or lack thereof) > sensor size > sensor pixel type > readout speed > sensor-to-AD-convertor path, A/D convertor (both bit depth and quality) > heat sinking / cooling > raw file compression > black calibration > in camera raw data manipulation > characteristic curve > ICC profile > demosaic algorithm > deconvolution algorithm > noise reduction type > up-res or down-res algorithm > sharpening
Any one of the above can influence the final image. It is a system, and no one component is as important as the overall system.
My camera uses a 16-bit ADC. Perhaps under ideal conditions that's irrelevant but under less-than-ideal conditions when I have to push the files' limits it falls apart much less often than files from 12- and 14-bit cameras. Would your left brain like to explain this?
You have failed to isolate the variables, a typical failure of a non-scientific approach. You should ask, "what other variables contribute to the observed robustness of the files". As pointed out earlier, photon noise and read noise have a lot to do with clean shadows. The newer generation of dSLRs such as the Nikon D7000 and Pentax D5 have very low read noise and very clean shadows, but are limited by photon noise because of their small sensor size. Photon noise predominates except in the deepest shadows. On the other hand, the MFDBs have a large sensor area allowing collection of more photons and a better SNR from photon noise, but are handicapped by high read noise. A noise analysis using photon noise and read noise such as in Table 2 of Roger Clark's (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html) treatise can describe the noise performance of a sensor fairly well, but other factors such as pattern noise and unfavorable coefficients for the 3x3 matrix transform as referenced by Emil in an earlier post come into play as well.
When micro-contrast and image detail enter into the equation, many more variables are involved.
Regards,
Bill
Yes, Doug (Telyt) takes some wonderful bird and wildlife pictures with his Leica/DMR.
I had a DMR and can also attest to its fantastic color and detail. I think my post where I measured the DR of the DMR and Canon 5D using Imatest and a stouffer transmission test wedge can still be found on these forums. In terms of DR, I don't think the DMR would equal the current cameras but it did well then.
One thing I've always felt is that people measure DR from light to dark, but there should be some kind of way to measure the camera's ability to reach across the colors similarly. Some cameras are able to show subtle changes in colors, and others not. The DMR was one of those that could render subtle color transitions very well. I've always felt this was something the Kodak CCD sensors were very good at and have always right or wrong credited that ability to the 16 bit A/D pipeline.
Yes probably the color has something to do with the filter array. But also I am betting in addition to that good A/D electronics also help a camera render those subtle color transitions, and that those cameras with higher bit pipelines can do it better.I think that it is often sufficient to have a model that considers the SNR/DR/Noise/non-linearity of luminance, and a separate model that considers color-characteristics as a linear, noise-free function of wavelength. This model certainly breaks down when considering raw development, but I think it is a good one for analyzing sensor/camera/raw behaviour.
Yes probably the color has something to do with the filter array. But also I am betting in addition to that good A/D electronics also help a camera render those subtle color transitions, and that those cameras with higher bit pipelines can do it better.
As far as I know "Telyt" is using a Leica DMR so it's a 1.3 crop factor digital back with a Kodak CCD. I don't know the pixel size. I also know that "Telyt" takes very good pictures, so whatever the bits, the camera serves him well.
The S2 is a very good camera and Telyt is said to take excellent pictures. However, his image quality is likely related to factors other than the bit depth of 16. Your DR analysis of 11 - 12 stops would require a bit depth of around 12 bits for encoding. One can reach similar conclusions through a noise model similar to that used by Roger Clark (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html) where the two main sources of noise--shot noise and read noise--can be added in quadrature to obtain total noise. As Emil (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#bitdepth) has explained, it makes little sense to quantize the signal from the sensor in steps much finer than the level of the noise.
The Leica S2 uses the KAF 37500 (http://www.kodak.com/ek/US/en/Image_Sensor_Solutions/KODAK_CCD_Image_Sensors_Power_New_Cameras_for_Professional_Photography.htm) sensor which was designed specifically for the S2, and Kodak has not released a data sheet. However, they do state that the chip uses the 6.0 micron TrueImage technology, and the performance is probably similar to other chips in this series. The KAF-40000 (http://www.kodak.com/ek/uploadedFiles/Content/Small_Business/Images_Sensor_Solutions/Datasheets%28pdfs%29/KAF-40000ProductSummary.pdf) is one of these chips and it has a full well of 42K electrons and a read noise of 13 electrons. The DR is listed at 70.2 db (11.7 stops), in line with your estimate.
The chart below shows the Clark style noise model along with the sensor gain (electrons per data number [DN]) for various bit depths, assuming that the full range of the ADC is utilized. At a SNR of 1, total noise is 13 electrons and even at a bit depth of 12, the gain is 10.25 electrons/DN. This meets Emil's criterion. A bit depth of 14 would give a margin of error, but a bit depth of 16 serves only to quantify noise.
Regards,
Bill
The S2 is a very good camera and Telyt is said to take excellent pictures. However, his image quality is likely related to factors other than the bit depth of 16. Your DR analysis of 11 - 12 stops would require a bit depth of around 12 bits for encoding. One can reach similar conclusions through a noise model similar to that used by Roger Clark (http://www.clarkvision.com/imagedetail/evaluation-1d2/index.html) where the two main sources of noise--shot noise and read noise--can be added in quadrature to obtain total noise. As Emil (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#bitdepth) has explained, it makes little sense to quantize the signal from the sensor in steps much finer than the level of the noise.
The Leica S2 uses the KAF 37500 (http://www.kodak.com/ek/US/en/Image_Sensor_Solutions/KODAK_CCD_Image_Sensors_Power_New_Cameras_for_Professional_Photography.htm) sensor which was designed specifically for the S2, and Kodak has not released a data sheet. However, they do state that the chip uses the 6.0 micron TrueImage technology, and the performance is probably similar to other chips in this series. The KAF-40000 (http://www.kodak.com/ek/uploadedFiles/Content/Small_Business/Images_Sensor_Solutions/Datasheets%28pdfs%29/KAF-40000ProductSummary.pdf) is one of these chips and it has a full well of 42K electrons and a read noise of 13 electrons. The DR is listed at 70.2 db (11.7 stops), in line with your estimate.
The chart below shows the Clark style noise model along with the sensor gain (electrons per data number [DN]) for various bit depths, assuming that the full range of the ADC is utilized. At a SNR of 1, total noise is 13 electrons and even at a bit depth of 12, the gain is 10.25 electrons/DN. This meets Emil's criterion. A bit depth of 14 would give a margin of error, but a bit depth of 16 serves only to quantify noise.
Regards,
Bill
Having the Hasselblad H4D-40 and Pentax 645D the same Kodak KAF-40000 sensor, how come Pentax state 14 bits and Hasselblad 16 bits ??
Whatever the reason would it bring better tonality to the Hasselblad?
Regards
ACH
As far as I know "Telyt" is using a Leica DMR so it's a 1.3 crop factor digital back with a Kodak CCD. I don't know the pixel size. I also know that "Telyt" takes very good pictures, so whatever the bits, the camera serves him well.
Erik, thank you for your reply..