Real World Dynamic Range

[NikoJorj]

Time to insert a link about noise, bit depth, dynamic range and S/N ratio...
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html
Don't forget to read the 2 previous pages about noise.

[ErikKaffehr]

Hi,

Yes, Emil's article is a very good read. I would just add a couple of observations.

Mark Dubovoy mentioned the term "full texture" in connection with dynamic range. It was suggested on another discussion that MTF may have to do with it. The way I see it, dynamic range on it's own is pretty useless. What we often want to see is some resolved detail. Now, if we have the same field of view, any detail will be like twice the size on MF compared to FX (full frame 135). So if we compare MTF curves we should compare MTF at 80 lp/mm for FX with MTF for 40 lp/mm on MF. There is no question that MF will be able to transfer much more contrast on tiny details than FX.

Another observation I may have is that I compared well shot raw images from the Leica S2 with equally well shot images from a Nikon D3X. The Leica images were sharper, no doubt about that, but the Nikon did in my view surpass the Leica S2 regarding DR. Check the end of this article: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/38-observations-on-leica-s2-raw-images

On the other hand I also have downloaded a couple of comparison images shot by Phase One P65+ and Canon 1DsIII from Phase One website. In this case the Phase images were much cleaner at low ISO. That later test did indicate a significant advantage of the P65+ over Canon 1DsIII.

One of the issues I see is that Phase One is also developing Capture One. Capture One may utilize information not available to other software vendors like very exact calibration data embedded in the images. I don't think this is an unfair advantage for Phase users, but I could say that they are fortunate to have very good backs with very good software and access to proprietary data.

Best regards
Erik

Time to insert a link about noise, bit depth, dynamic range and S/N ratio...
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html
Don't forget to read the 2 previous pages about noise.

[Doug Peterson]

The way I see it, dynamic range on it's own is pretty useless. What we often want to see is some resolved detail.

+1

The engineering/scientific definition(s) of dynamic range are never my concern. My question is always how do the shadows/highlights LOOK, especially when pushed/pulled. Are tonal transitions smooth or choppy? If there is noise is it random and pleasing grain or it is clumpy and broken up and ugly? Are the colors accurate especially when the same subject has a shadow and highlight on it does the color remain consistent even after a push or pull into the deep shadows or hot highlights? What happens at the breaking point? Is there a hard edge where there is no more data or does it fade away gently into nothingness?

Which transitions into the second point:

One of the issues I see is that Phase One is also developing Capture One. Capture One may utilize information not available to other software vendors like very exact calibration data embedded in the images. I don't think this is an unfair advantage for Phase users, but I could say that they are fortunate to have very good backs with very good software and access to proprietary data.

That's the value of a system. The software development team and the hardware development team at the Denmark headquarters of Phase One are about 50 meters apart and not only work together but drink together, hang out together, and share the same passion for making the best possible product. They've also been working on eeking the most out of the system of hardware/firmware/software since 1997. In that regard I'm always surprised when people ignore the value of having a dedicated raw developer. These guys obsess about how they can make the SYSTEM perform best in photographic terms.

Doug Peterson (e-mail Me)
__________________
Head of Technical Services, Capture Integration
Phase One, Leaf, Cambo, Canon, Apple, Profoto, Eizo & More
National: 877.217.9870  |  Cell: 740.707.2183
Newsletter: Read Latest or Sign Up
RSS Feed: Subscribe
Buy Capture One at 10% off
Personal Work

[Nick Rains]

Erik

I have done the S2 vs D3X test myself and almost agree with your findings. I found the D3X had subtly better DR but the deepest shadows had a lack of texture that I put down to some NR going on in the camera - on the NEF. I coud be wrong, but I do see a slight smudgy plastic look in the deep tones that looks very much like NR. The S2 is also slightly cleaner when you really push the tones around.

One a pixel level I'd call it a draw, both have 6um pixel pitch and appear to resolve the same details over the same sensor area. The S2 only leads when you take into account the extra sensor area.

The D3X is a truly amazing camera and those new Nikon lenses are sweet.

I fully agree about the P65+ vs 1Ds MkIII, I'm a Canon user too, and the current top of the line sensor is no match for the D3X or any of the MFDBs. I've not tried the P65+, but I have used the P45+ and that it still quite amazing.

I tried to massively pull out the shadows in my 9 stop 'real world' test yesterday from my 5D2 and there was a world of difference from the S2.

I also agree that DR as it stands is not useful unless there is also contrast between recorded details - that what makes top prints stand out, smooth tones and fine detail all the way from max black to max highlights.

[ErikKaffehr]

Hi,

A couple of issues. A sensel (sensor cell) can hold a certain number of electrons, with each electron representing a captured photon (by and large). A sensel may hold something like 50000 electrons. The electrons will be interpreted as a voltage by the ADC (Analog Digital Converter). Assuming that we would have a Full Well Capacity of 50000 electrons a 16 bit ADC would be able to resolve all possible combinations of electrons, while a 12 bit converter would only be able to resolve 4096 different combinations.

Now, in real life we have noise. That noise is coming from the readout circuitry but also from the distribution of the photons themselves. If we have like 10 electrons of readout noise the number of meaningful combinations of electrons will be around 5000, which is quite close to 12 bits (4096 values).

Aside from that we also have what is known as shot noise, random variations in the number of incident photons. The photons are Poisson distributed which leads to the noise being the square root of the signal. So with 100 photons captured the signal would vary between 90 and 110 electrons.

The final factor is exposure. The full well capacity is only utilized fully if we expose correctly to the right and we don't know if we ever achieve this. The histogram on the camera is only an estimate of what the sensor sees.

To sum it up, 16 bit capture is enough to detect each electron in the sensel of todays sensors, but the last 3-4 bits will essentially just measure noise. Whatever the electronics, some noise will always remain as it is also a characteristic of the discrete nature of light.

Best regards
Erik

Thanks Mike, that makes sense - I was about to agree with Christopher but this clears it up. You are saying that, in the absence of noise, a 16 bit sensor can always record 16 stops (doublings) of light intensities - by definition. The usable DR is then limited by signal noise levels.

Does this sound about right?

[ondebanks]

I found the D3X had subtly better DR but the deepest shadows had a lack of texture that I put down to some NR going on in the camera - on the NEF. I coud be wrong, but I do see a slight smudgy plastic look in the deep tones that looks very much like NR.

Interesting! But that wouldn't surprise me. Nikon already "have form" (m'lud) for screwing around with (filtering) their RAW files in long exposures. NEF most definitely is not a true, straight-off-the-sensor RAW format. Perhaps now they're extending the crime to filtering deep tones in short exposures as well. 
DxOlabs showed conclusively that Sony are also guilty of RAW filtering - at high ISO:
http://dxomark.com/index.php/en/Our-publications/DxOMark-Insights/Half-cooked-RAW/Noise-reduction

Why call it RAW when it isn't? I think the Advertising Standards Authority should prosecute Nikon and Sony for misleading portrayal of their products!

I fully agree about the P65+ vs 1Ds MkIII, I'm a Canon user too, and the current top of the line sensor is no match for the D3X or any of the MFDBs.

Except for long exposures and high ISO, where you would have to drop "or any of the MFDBs" from that sentence.

Ray

[dudu307]

Interesting! But that wouldn't surprise me. Nikon already "have form" (m'lud) for screwing around with (filtering) their RAW files in long exposures. NEF most definitely is not a true, straight-off-the-sensor RAW format. Perhaps now they're extending the crime to filtering deep tones in short exposures as well.  
DxOlabs showed conclusively that Sony are also guilty of RAW filtering - at high ISO:
http://dxomark.com/index.php/en/Our-publications/DxOMark-Insights/Half-cooked-RAW/Noise-reduction

Why call it RAW when it isn't? I think the Advertising Standards Authority should prosecute Nikon and Sony for misleading portrayal of their products!

Except for long exposures and high ISO, where you would have to drop "or any of the MFDBs" from that sentence.

Ray

I don't get your point.

First, DXO didn't find filtering in D3x. Second, what's the big deal in offering a true RAW but then force you to use proprietary software that use proprietary calibration files to hide the problems of your sensor.

If you use a more "straight" raw developer (like Raw Photo Processor) D3x files always look great, medium format files sometimes look amazing, sometimes look like crap..... but yes you don't have the great Capture one to hide all the problems....

At the end of the day, I want the best possible raw file, and use my raw converter of choice. I don't want hidden CCD problems, hidden calibration files and hidden fixes in software.

[madmanchan]

It depends on the camera model. Some cameras will apply calibration data and do other processing in the camera, prior to writing out the raw file. Other cameras will leave it to software (i.e., desktop raw conversion software, not in the camera) to apply the calibration data later. So maybe you can think of "raw" terminology in terms of "raw", "rare", "medium rare", etc.   

[Doug Peterson]

First, DXO didn't find filtering in D3x. Second, what's the big deal in offering a true RAW but then force you to use proprietary software that use proprietary calibration files to hide the problems of your sensor.

If you use a more "straight" raw developer (like Raw Photo Processor) D3x files always look great, medium format files sometimes look amazing, sometimes look like crap..... but yes you don't have the great Capture one to hide all the problems....

At the end of the day, I want the best possible raw file, and use my raw converter of choice. I don't want hidden CCD problems, hidden calibration files and hidden fixes in software.

Incorporating the raw processor into the image quality chain means that as math improves you can reprocess any given raw file and get better quality. An H25 (2003 vintage) looked great when it came out, but looks even better now when processed through Capture One 5. Color, noise, dynamic range, tonal smoothness, lens corrections (chromatic aberration, distortion removal) have all improved considerably in that time. The more a company decides to bake the raw file in-the-camera rather than in the software the more you lose that very important advantage. For day to day work it's probably not important, but for the best 20 shots you ever take in your life I would MUCH rather have a completely and utterly naked raw file with any corrective algorithms placed in the software. Just my opinion of course.

But other than wide-angle technical cameras (where LCC is needed) you are not "forced" into using Capture One with a Phase One / Leaf / Mamiya back. The files are supported in ACR (Lightroom/Photoshop), and Irident Raw Developer (great app if you've never used it), and many in Apple Aperture and others, and if you insist on going for compatibility over quality you can also export to DNG and use them with just about anything. It's just that most of our users of high-end digital backs don't want to buy an expensive camera and then use a raw processor that gives them 2nd-best results.

By the way - have you processed a difficult (long exposure, high ISO, mixed lighting, underexposed, etc) Nikon D3X through both LightRoom and Capture One 5.2? You might change your mind about the whole issue :-).



Doug Peterson (e-mail Me)
__________________
Head of Technical Services, Capture Integration
Phase One, Leaf, Cambo, Canon, Apple, Profoto, Eizo & More
National: 877.217.9870  |  Cell: 740.707.2183
Newsletter: Read Latest or Sign Up
RSS Feed: Subscribe
Buy Capture One at 10% off
Personal Work