Nikon D7000 Dynamic Range

[Ray]

It is interesting to see how digital photography gets better and better. I just used a Nikon d3s and was amazed how much colour it sees in almost total darkness using a 1.4 lens wide open.
My eyes are less sensitive! and see for sure less colour.
( and i am not blind at all)

Indeed! I find the low-light capability of modern cameras very useful. Taking photos in low light situations using a Speedlite or built-in flash can produce unwanted results such as black shadows around edges, and total darkness prevailing just a few metres away because of the limited range of the flash.

Flash can also be annoying for the subject and can bring unwanted attention to oneself. Some time ago whilst photographing a cabaret show in Bangkok, whilst sat in the first row, the manager approached me and asked me to stop using flash because it was distracting the performers. Of course I complied and for the rest of the show had my 5D set on ISO 1600 & 3200. Frequently the DoF was not sufficient or the shutter speed too slow for a sharp shot consistent with reasonably acceptable noise.

In that situation, a D3s would have been ideal, but that camera's a bit too heavy and expensive for me. A D7000 would not have helped much in that situation, except in respect of a more extensive DoF at the same aperture and shutter speed.

[stamper]

Ray .... the men in white coats are coming. Don't resist them, they only want to help.

[BernardLanguillier]

The key thing I will retain from these discussions isn't so much how good the D7000 is or it not relative to the P65+. It is also not how close the DxO DR is to real DR.

What I retain, especially after reading fascinating threads at DPreview, is that most people - including people going pretty deep into the technicalities of photography - are not able to measure DR as part of their photographic experience. DR is most often confused with the ability to recover highlights (a property that only depends on sensor ISO calibration) while it is only about noise in under-exposed areas.

Designing a sensor that will have very clean shadows involve making other compromises in order to achieve a general balance. This involves huge R&D investments and is probably an order of magnitude more complex than raising pixel count. Yet most customers don't see it and keep fighting every objective measure of it.

We - more or less advanced photographers - have been just as guilty as the rest. We have been saying "more DR", but are mostly unable to recognize it when it shows up.

The net result of this is most probably going to be next generation sensors with less DR... why should these companies keep investing huge money to improve a camera characteristic that we think we care about, but actually don't?

They would all be much better off do what Phaseone does, under-expose all their shots one stop by turning ISO 100 into ISO 200, and calibrate the software chain so as to be able to recover one stop of highlights. So much so for the myth that advanced photographers prefer to shoot with slides films. 

Cheers,
Bernard

[Ray]

The key thing I will retain from these discussions isn't so much how good the D7000 is or it not relative to the P65+. It is also not how close the DxO DR is to real DR.

What I retain, especially after reading fascinating threads at DPreview, is that most people - including people going pretty deep into the technicalities of photography - are not able to measure DR as part of their photographic experience. DR is most often confused with the ability to recover highlights (a property that only depends on sensor ISO calibration) while it is only about noise in under-exposed areas.

Designing a sensor that will have very clean shadows involve making other compromises in order to achieve a general balance. This involves huge R&D investments and is probably an order of magnitude more complex than raising pixel count. Yet most customers don't see it and keep fighting every objective measure of it.

We - more or less advanced photographers - have been just as guilty as the rest. We have been saying "more DR", but are mostly unable to recognize it when it shows up.

The net result of this is most probably going to be next generation sensors with less DR... why should these companies keep investing huge money to improve a camera characteristic that we think we care about, but actually don't?

They would all be much better off do what Phaseone does, under-expose all their shots one stop by turning ISO 100 into ISO 200, and calibrate the software chain so as to be able to recover one stop of highlights. So much so for the myth that advanced photographers prefer to shoot with slides films. 

Cheers,
Bernard

Those are a good points, Bernard.

Speaking for myself, the only reason I'm concerned about issues of DR, is because in the past I've sometimes been disappointed with the quality of shadow detail, not only regarding shadow detail in certain slides I've scanned, but also the shadow detail at base ISO from DSLRs I've used.

In the house of a friend of mine, there's a very large and impressive print of a waterfall hanging on the wall, taken by a professional photographer who appears to have done all the right things, neutral density filter for a slow shutter speed, and polariser to smooth the specral reflections on the water surface at the foot of the falls.

But in my view, the scene has one major flaw. All the deep shadows are totally black (0,0,0,).

From a distance, that's not too bad, but from close-up it looks a bit odd to me, a bit unnatural.

Would it be fair to presume that the reasons such shadows are totally black is not just for an esthetic reason, black is beautiful, but to hide the fact that the detail in those shadows is crap, due to a lack of sufficient DR capability of the camera and the processing chain?

Reluctant as I am to show my rejects in order to demonstrate a point, here are a couple of casual shots of a waterfall in a rainforest taken several years ago with a 6mp DSLR and Sigma zoom at base ISO. They were casual, no tripod or ND filter, but correctly exposed in accordance with the principles of ETTR. The 100% crops highlight the problems in the deep shadows.

Another interesting characteristic of the D7000, which has been noticed by some perceptive posters at dpreview, is that DR is now so good at base ISO, there is no image-quality advantage in using a high ISO setting.

In other words, the D7000 now behaves like a DB. An ISO 800 shot is no better than an ISO 100 shot underexposed 3 stops.

There are both advantages and disadvantages to this situation, ie, underexposing at base ISO instead of increasing ISO for the desired shutter speed at the desired aperture.

The disadvantage is that the review jpeg on the camera's LCD might be too dark for you to properly assess if you've captured the moment successfully, although I tend to think this is a trivial concern. If you havn't captured the moment successfully, too bad. The moment has passed.

For me, I think the main purpose of the LCD review is to determine whether I've blown the highlights, which brings me to the advantages of this characteristic of the D7000.

The concept of ETTR (expose to the right) has been a much-discussed topic on this forum over the years. So much so that the issues should now be clear to everyone.

If you do not give a 'full' exposure, whatever the ISO setting, you are not going to get the full qualities of tonal range and low noise that your camera is capable of.

However, there's a huge practical problem with this methodology of ETTR. In one's attempt to get a 'full' exposure, it's just too easy to blow highlights.

If you have the time, and can retake the shot because the 'moment' is not critical, the LCD review image is useful, at base ISO, even on the D7000.

However, if you are not as base ISO because you need a faster shutter speed and/or need to stop down for greater DoF, then the risk of blowing highlights is of greater concern.

This quality of the D7000 seems to me to be a significant advantage for 'manual' shooters. You choose the aperture and shutter speed that's required for the shot, in the circumstances, whatever the lighting. If you raise ISO, you have to worry about ETTR issues. If you leave the camera at base ISO, the image will be underexposed, but you can be completely confident there will be no blown highlights and no loss of image quality, when shooting RAW of course, after compensating for the underexposure in ACR.

I like it.

[BernardLanguillier]

Speaking for myself, the only reason I'm concerned about issues of DR, is because in the past I've sometimes been disappointed with the quality of shadow detail, not only regarding shadow detail in certain slides I've scanned, but also the shadow detail at base ISO from DSLRs I've used.

We have been screaming for more DR, but my point is that we are not always being super coherent on this one. The logical consequence should be less focus on DR from our camera manufacturers. As always there is strong inertia between consumer demand and product development so I believe that next gen products - speced one or 2 years ago - will still show an improvement, but my guess is that the following generation will see either a stagnation or a decrease of DR since few people really care.

You'd think that all landscape, architecture, product,... photographers would have the same concerns, but this doesn't appear to be the case, or at least to be enough the case to translate into significantly added sales for those companies having invested huge amounts of money in really improving DR:

- Some photographers really don't care, they focus on the same scenes they have been focusing on since slides days and just ignore the other scenes. You don't need DR then and they might be right, what is nicer than a well exposed slide with sweet light hitting the very right spot in a composition?
- Some do care and are using these cameras/backs delivering more real world DR,
- Some do care but they don't know that there is something better than their existing camera in terms of shadow noise,
- Some do care, know there is something better than their existing camera but don't see the value proposition of DR as being good enough to overcome the cost of a switch now (heck, other brands will catch up some day),
- Some do care, know there is something better than their existing camera but won't admit it because of some crazy self defeating fanboy behavior or whatever even less admissible reasons,
- ...

I am NOT thinking of any person in particular.

Cheers,
Bernard

[Ray]

We have been screaming for more DR, but my point is that we are not always being super coherent on this one. The logical consequence should be less focus on DR from our camera manufacturers. As always there is strong inertia between consumer demand and product development so I believe that next gen products - speced one or 2 years ago - will still show an improvement, but my guess is that the following generation will see either a stagnation or a decrease of DR since few people really care.

Not sure about that, Bernard. What we seem to have now are, on the one hand, cameras that have brilliant DR at low ISO, such as the D3X and D7000, but modestly good DR at high ISO, and on the other hand, cameras that have brilliant DR at high ISO, such as the D3s, but only modestly good DR at low ISO.

Hopefully, DSLRs of the future, whether Nikon or Canon, will combine these two strengths so we will have the best of both worlds. 

[BernardLanguillier]

Not sure about that, Bernard. What we seem to have now are, on the one hand, cameras that have brilliant DR at low ISO, such as the D3X and D7000, but modestly good DR at high ISO, and on the other hand, cameras that have brilliant DR at high ISO, such as the D3s, but only modestly good DR at low ISO.

Hopefully, DSLRs of the future, whether Nikon or Canon, will combine these two strengths so we will have the best of both worlds. 

That will only happen if they feel their customers care and it helps selling more cameras at a higher price point.

Cheers,
Bernard

[ErikKaffehr]

Hi!

Whats your main wish? Better DR or usable histograms?

There are two ways to increase DR, one is to increase full well capacity, the other is to reduce read noise. The second option improves DxO ratings but may have little effect in normal photography which is dominated by shot noise, at least in my humble opinion.

Usable histograms would help us to really expose to the right and improve photon statistics.

The look of the noise also matters, some noise is worse than others.

Best regards
Erik

That will only happen if they feel their customers care and it helps selling more cameras at a higher price point.

Cheers,
Bernard

[RFPhotography]

Not only farting around with dynamic range tests, but farting around reading books, and farting around photographing temples at night with only street lighting for illumination, and being concerned about unacceptable noise in the shadows; and farting around getting to know the limitations of my equipment, sometimes processing images in the evening on my laptop to see how the day's shooting went and to see what mistakes I made, in the hope of not repeating them.

A holiday which allows for no time to fart around is no holiday in my opinion. My last trip was an organised cruise along rivers in Europe and Russia. Everyday there were organised tours from morning till dusk, then yet more tours in the evenings to see shows, Russian folk dancing, ballets, concerts etc. By the time I got back to Australia, I felt so exhausted I needed a holiday.

Get my point?

I do get your point.  It seems you're missing mine; however.  As an aside, organised tours aren't my cup of tea either.  I prefer to find and explore on my own.

Erik summed it up quite nicely in his last comment.  Of course I want to understand the limits of my equipment (and I do) but I'm not going to waste time replicating DxO or DPReview or anybody else's tests to try and do it.  And I'm certainly not going to set out to try to prove DxO or other test sights wrong.  And most certainly not going to do it while on vacation.  My interests in the limitations of my gear are around practical uses.  Like many, I'll research new gear by looking at what I feel are credible reviews but those only form part of the buying decision and a minority part at that.  The larger part of the decision making process will come from the real world experiences of people whose opinions I value and my own experience from renting/borrowing before buying or taking a memory card to the store and firing of a bunch of shots in different conditions in store and bringing those back to my office to evaluate.  I've actually been asked to leave a store because the salesperson felt I was taking too many shots with their demo camera.  Apparently he worked on the look but don't touch theory of selling.   

I also understand that, as I noted earlier, technology improves.  It's not a shock when new equipment is better than previous generation equipment.  It is a disappointment when new is not as good as prior.  It's also a serious disappointment; and a larger one than with respect to the technological aspects, when serious design, firmware programming and manufacturing QC problems plague a manufacturer.  If a camera fails in the field it doesn't matter how good its DR is.  If you can't focus a camera in the field, it doesn't matter how good its DR is.  If a camera's mirror falls off in the field it doesn't matter how good its DR is.

So, in conclusion, I'm far more concerned with the practical aspects of what a camera is capable of than the theoretical. 

[bjanes]

The concept of ETTR (expose to the right) has been a much-discussed topic on this forum over the years. So much so that the issues should now be clear to everyone.

If you do not give a 'full' exposure, whatever the ISO setting, you are not going to get the full qualities of tonal range and low noise that your camera is capable of.

However, there's a huge practical problem with this methodology of ETTR. In one's attempt to get a 'full' exposure, it's just too easy to blow highlights.

If you have the time, and can retake the shot because the 'moment' is not critical, the LCD review image is useful, at base ISO, even on the D7000.

However, if you are not as base ISO because you need a faster shutter speed and/or need to stop down for greater DoF, then the risk of blowing highlights is of greater concern.

This quality of the D7000 seems to me to be a significant advantage for 'manual' shooters. You choose the aperture and shutter speed that's required for the shot, in the circumstances, whatever the lighting. If you raise ISO, you have to worry about ETTR issues. If you leave the camera at base ISO, the image will be underexposed, but you can be completely confident there will be no blown highlights and no loss of image quality, when shooting RAW of course, after compensating for the underexposure in ACR.

Ray,

Excellent points. By extension of your thoughts, I propose that exposure to the right (ETTR) really applies only to base ISO. If you expose at base ISO so that the highlights are just short of clipping and then double the camera ISO, the sensor will only be at half saturation (full well), but the electronic gain will be increased so that the histogram is moved to the right and you have a nice looking histogram. However, the shot noise will be the same as with the histogram of the "underexposed" shot at base ISO.

With the old generation of cameras such as the D3 and D3s, the increase in ISO reduced read noise and improved the SNR in the shadows. However, with the newer generation cameras with improved electronics and ADCs, read noise is optimal at base ISO and the use of a higher ISO results in less headroom for the highlights and can result in burnt highlights. Even with older cameras, one can increase ISO only to the point where read noise approaches a minimum (easily predicted by the DXO DR curves as Emil explains in the above quoted link. For the Nikon D3, this occurs at about ISO 600 where the DR curve becomes linear.

Regards,

Bill

[Ray]

I do get your point.  It seems you're missing mine; however.  As an aside, organised tours aren't my cup of tea either.  I prefer to find and explore on my own.

Erik summed it up quite nicely in his last comment.  Of course I want to understand the limits of my equipment (and I do) but I'm not going to waste time replicating DxO or DPReview or anybody else's tests to try and do it.  And I'm certainly not going to set out to try to prove DxO or other test sights wrong.  And most certainly not going to do it while on vacation.  My interests in the limitations of my gear are around practical uses.  Like many, I'll research new gear by looking at what I feel are credible reviews but those only form part of the buying decision and a minority part at that.  The larger part of the decision making process will come from the real world experiences of people whose opinions I value and my own experience from renting/borrowing before buying or taking a memory card to the store and firing of a bunch of shots in different conditions in store and bringing those back to my office to evaluate.  I've actually been asked to leave a store because the salesperson felt I was taking too many shots with their demo camera.  Apparently he worked on the look but don't touch theory of selling.   

I also understand that, as I noted earlier, technology improves.  It's not a shock when new equipment is better than previous generation equipment.  It is a disappointment when new is not as good as prior.  It's also a serious disappointment; and a larger one than with respect to the technological aspects, when serious design, firmware programming and manufacturing QC problems plague a manufacturer.  If a camera fails in the field it doesn't matter how good its DR is.  If you can't focus a camera in the field, it doesn't matter how good its DR is.  If a camera's mirror falls off in the field it doesn't matter how good its DR is.

So, in conclusion, I'm far more concerned with the practical aspects of what a camera is capable of than the theoretical. 

Hhmm! I'm not sure you did get my point. All my tests are done for practical reasons as an aid to best use my equipment, or to help me select a camera. The tests I quoted in Chiang Mai were done before DXOMark existed and were carried out to help clarify some confusion between absolute or engineering DR specifications, and useful DR that is subjectively acceptable.

On a later trip to Thailand, I also spent the equivalent of a whole day comparing the new Nikon D3 with my old 5D at the main Nikon store in Bangkok. That test was also done before DXOMark existed. I did that in order to check for myself the incredible claims being made at the time that the high-ISO performance of the D3 was up to 2 stops better than any other DSLR on the market. If such claims were true, I wanted one. Alas! They weren't true. At ISO 1600 and above, I wasn't able to see more than 1/2 a stop improvement

Perhaps you do not understand that I'm retired and am therefore on holiday wherever I happen to be, in Australia or in Thailand. The cost of living in Thailand is lower than Australia. I sometimes think I would be financially better off if I were to live there permanently. I might then be able to afford expensive cameras. 

[Ray]

Hi!

Whats your main wish? Better DR or usable histograms?

There are two ways to increase DR, one is to increase full well capacity, the other is to reduce read noise. The second option improves DxO ratings but may have little effect in normal photography which is dominated by shot noise, at least in my humble opinion.

Usable histograms would help us to really expose to the right and improve photon statistics.

The look of the noise also matters, some noise is worse than others.

Best regards
Erik

You might be right Erik, but I'm a bit doubtful that we've reached that stage of development where all parts of the image, in all circumstances, are dominated by shot noise in the latest cameras such as the D7000 and K-5.

Even if it is true, that shot noise dominates, that does not necessarily mean there is no room for improvement, does it? For example, if total noise at a particular signal level consists of 55% shot noise and 45% readout noise, including other noise, such as traces of dark-current and fixed pattern noise, then one could claim that shot noise dominates.

However, if we reduce the 45% component to 25%, for example, we will have increased dynamic range, is that not so?

Another issue, about which I admit I'm a bit hazy, is the quantum efficiency of the silicon photo-diodes, or pixels. As I understand, the sensels gather a charge as a result of photons knocking off electrons in the silicon. If a pixel has a full-well capacity of 30,000 electrons, the quantum efficiency will determine how many photons were required to create that charge. If the pixel has a QE of 50%, then 60,000 photons would be required to create that maximum full-well charge of 30,000 electrons. Is this not so? I confess, I'm no expert in this area, but this is how I understand it. Perhaps Bill Janes can comment.

If the the QE of the sensel is increased, even if all other factors remain the same, the useable DR of the sensor will also be increased, will it not?

As I see it, shot noise (or its causes) exists before the image is captured. It's part of the fabric of reality. The perfect camera would capture an image with no less than its full quota of shot noise, in all its glory. Shot noise is the only noise we want to see. That's reality.

[BJL]

If the DXo results are right about the SNR ratio hitting 1:1 only when one gets a full 14 stops below full well capacity on some recent DSLRs, then for all practical purposes, shot noise in the incoming light is the only noise source of practical significance for such sensors, because read noise has been reduced to a level that is insignificant. Note to Ray: "insignificant" is different from "exactly zero".

Why? Because
1. It seems fairly well established (in all but the obsessive pixel/lab test peeping community) that for normal photographic purposes -- as opposed to aerial mapping, machine inspection, security surveillance, medical imaging and so on -- parts of the image with SNR ratio below 10:1 are rather poor, and below about 5:1 is garbage that should be left hidden in the deep shadows of the final displayed image. And to get SNR due to shot noise alone down to 5:1 requires at least 5^2 photons detected, or 25e- of signal. (There is no hint of more than 1e- per detected photon in these sensors, so I will assume detected photons = photo-electrons from now on. More than 1e- per photon would make the dominance of shot noise over read noise even greater.)
2. The well capacity of the photosite sizes in those sensors cannot be more than about 30,000e-: it if were, base ISO speed would be lower than it is. So 14 stops below full is a signal of no more than about 30,000e-/2^14, about 2e-, and if SNR=1 at that level, the read noise is no more than about that same 2e-. (This fits with measurements from other sources of under 3e- for some recent Canon sensors with larger photosites.)
3. For signal of less than 25e-, the results are garbage that needs so to stay in very dark parts of the displayed image, and nothing much can be done about that.
4. At 25e- signal, shot noise is 5e, and adding at most 2e of read noise in RMS fashion gives at most sqrt(5^2+2^2) = 5.4, reducing SNR by less than 10% from 5:1 to 4.63:1: from "bad" to "very slightly worse". This is very unlikely to have any significant visual consequences!
5. As you move above 25e- to more useful signal levels, the contribution of 2e- or less of read noise becomes even less significant. For example, to get the 10:1 local SNR often described as minimally acceptable requires at least 100e- of signal, and adding 2e- of read reduces the SNR only 2%, from 10:1 (shot noise only) to 9.8:1 (shot + read).

Given the evidence from tests on other (Canon CMOS) sensors suggesting comparably low read noise levels (<2e-), and given that for some years special read-out methods in some technical devices has been capable of getting read noise below 1e- RMS, the idea of 2e- read noise is not at all crazy, so maybe these DXo results are for real.

But the "tonal range" measure at DXo seems far more relevant as a predictor of perceived image quality, and it is far lower: 9 stops or less.

[Ray]

If the DXo results are right about the SNR ratio hitting 1:1 only when one gets a full 14 stops below full well capacity on some recent DSLRs, then for all practical purposes, shot noise in the incoming light is the only noise source of practical significance for such sensors, because read noise has been reduced to a level that is insignificant. Note to Ray: "insignificant" is different from "exactly zero".

Why? Because
1. It seems fairly well established (in all but the obsessive pixel/lab test peeping community) that for normal photographic purposes -- as opposed to aerial mapping, machine inspection, security surveillance, medical imaging and so on -- parts of the image with SNR ratio below 10:1 are rather poor, and below about 5:1 is garbage that should be left hidden in the deep shadows of the final displayed image. And to get SNR due to shot noise alone down to 5:1 requires at least 5^2 photons detected, or 25e- of signal. (There is no hint of more than 1e- per detected photon in these sensors, so I will assume detected photons = photo-electrons from now on. More than 1e- per photon would make the dominance of shot noise over read noise even greater.)
2. The well capacity of the photosite sizes in those sensors cannot be more than about 30,000e-: it if were, base ISO speed would be lower than it is. So 14 stops below full is a signal of no more than about 30,000e-/2^14, about 2e-, and if SNR=1 at that level, the read noise is no more than about that same 2e-. (This fits with measurements from other sources of under 3e- for some recent Canon sensors with larger photosites.)
3. For signal of less than 25e-, the results are garbage that needs so to stay in very dark parts of the displayed image, and nothing much can be done about that.
4. At 25e- signal, shot noise is 5e, and adding at most 2e of read noise in RMS fashion gives at most sqrt(5^2+2^2) = 5.4, reducing SNR by less than 10% from 5:1 to 4.63:1: from "bad" to "very slightly worse". This is very unlikely to have any significant visual consequences!
5. As you move above 25e- to more useful signal levels, the contribution of 2e- or less of read noise becomes even less significant. For example, to get the 10:1 local SNR often described as minimally acceptable requires at least 100e- of signal, and adding 2e- of read reduces the SNR only 2%, from 10:1 (shot noise only) to 9.8:1 (shot + read).

Given the evidence from tests on other (Canon CMOS) sensors suggesting comparably low read noise levels (<2e-), and given that for some years special read-out methods in some technical devices has been capable of getting read noise below 1e- RMS, the idea of 2e- read noise is not at all crazy, so maybe these DXo results are for real.

But the "tonal range" measure at DXo seems far more relevant as a predictor of perceived image quality, and it is far lower: 9 stops or less.

Well first let me emphasise that I accept you know much more about physics and mathematics than I do, BJL, so I would hestitate to dispute your calculations.

But it seems to me we have a shortage of hard facts here, upon which to base our calculations. In such circumstances, practical experiments are the preferred option.

I did a Google search in an attempt to find out what the typical QE of modern sensors are, such as the 7D and D7000, and what the full-well capacity may be. Not much luck I'm afraid.

But I did come across a calculation that deduced the full-well capacity of the 7D pixel is 29,500e-. Whether or not that's accurate, I have no idea.

I also came across a Wikipedia entry containing a computer simulation of a series of images that contain only shot noise and nothing else. To quote:

A photon noise simulation, using a sample image as a source and a per-pixel Poisson process to model an otherwise perfect camera (quantum efficiency = 1, no read-noise, no thermal noise, etc).

Here's the link: http://en.wikipedia.org/wiki/File:Photon-noise.jpg  If you click twice on the image, you'll get a substantial enlargement.

Again, I've no idea if such a simulation is any more accurate than computer modelling that purports to demonstrate that increasing levels of anthropogenic CO2 levels are a serious threat to our well-being.

But for argument's sake, lets assume such simulations and estimates are true. Let's assume that the 29,500e- full-well capacity of the 7D is true. The D7000 with slightly larger photosites and a slightly larger sensor than the 7D, could then reasonably have a full-well capacity in excess of 30,000e-.

Let's call it 32,768e- for ease of calculation. If we start from a mean number of photons across the entire image of just one per pixel, at a given very short exposure, then double the exposure time in a series of 15 steps (1,2,4,8 etc), we arrive at 32,768 photons, after 15 doublings.

We have in effect a 16 EV dynamic range, presuming no additional noise from the camera. Is this correct?

Let's now refer to the Wikipedia simulation, assuming again it's a realistic simulation.

The first image on the second row, consisting of an average of 1 photon per pixel, is terribly noisy, but it's a recognisable image. If the subject matter were more interesting, a print of this image could hang in an art gallery.

"Ladies and gentleman, behold! The first 'one photon per pixel' digital image that has ever been recorded. Minimalism take a back seat."

(Darling! What the heck is he talking about. I don't understand a word of this.)"

[dimapant]

We probably all know by now that the DR of the D7000 at base ISO (together with that of the Pentax K-5) trumps every other digital camera on the market, whatever the format.

..........................

Are you really sure?
Did you print already an A2 side by side?

We will throw away all our stuff to get this miracle!

Many thanks for your information.

[Riaan van Wyk]

(Darling! What the heck is he talking about. I don't understand a word of this.)"

So say I.

[BJL]

Ray,
    first, thanks for confirming my rough estimate of about 30,000e-, and yes, I am happy to round up to 2^15=32,768. Which means that 14 stops below full well is 2^15/2^14 = 2 electrons, or two photons counted.

Second, it seems that I need to be careful when talking with you, in case I am misinterpreted. As I have said some time ago, one photon per pixel is just great for the original image recording, and with negligibly low read noise this would be the ultimate way to preserve spatial detail ... if you have enough pixels to dither heavily for display. Because this is exactly what standard B&W film does! It has many billions of photosites, silver halide crystals or clumps thereof, each of which either detects light or does not, and so provides a 1-bit output signal. To get acceptable visual results, you must then print or display at sufficiently high PPI/sufficiently large viewing distance, and so dither those extremely noisy individual pixels into an image that the eye detects as a lower resolution, higher DR, higher SNR image. (This is part of my refutation of the myth that a sensor with too many, too small pixels will have miserable IQ due to low DR: the confounding of per pixel DR and SNR with perceived noise levels.) I have done 100% pixel peeping of B&W film, by staring though a grain magnifying dark-room focusing aid, and the noise and DR ones sees then is atrocious ... but that says nothing about how the prints look, and not measure of a film's per pixel SNR or DR says anything about the DR of the prints.

In case anyone still misses my point, worrying about per pixel SNR and DR is nonsense without reference to pixel count, the PPI at which the image is displayed and to distance at which it is viewed, and such. People are getting "blinded by science" when trying to draw conclusions above IQ from DR measurements like those at DXo.


But I was talking about acceptable perceived noise levels after dithering, and the images you cite already have a lot of that, at least on the original page, as they are downsampled, so that each pixel on screen averages together multiple pixels from the original file, and has more than one photon per screen pixel. And do note the comment there:
Note the rapid increase in quality past 10 photons/pixel.
To which I would add that 10 photons per pixel is still quite nasty, 100 photons per pixel is a still noticeably grainy, like high ISO B&W film, while 1000 photons per pixel and up look fine to me. And at that nasty 10 photons per pixel and up, 2e- RMS of noise is insignificant, as noted in my previous post.

... Let's assume that the 29,500e- full-well capacity of the 7D is true. The D7000 with slightly larger photosites and a slightly larger sensor than the 7D, could then reasonably have a full-well capacity in excess of 30,000e-. ... Let's call it 32,768e- for ease of calculation.

... Let's now refer to the Wikipedia simulation, assuming again it's a realistic simulation.

The first image on the second row, consisting of an average of 1 photon per pixel, is terribly noisy, but it's a recognisable image. If the subject matter were more interesting, a print of this image could hang in an art gallery.

[Ray]

Ray,
    first, thanks for confirming my rough estimate of about 30,000e-, and yes, I am happy to round up to 2^15=32,768. Which means that 14 stops below full well is 2^15/2^14 = 2 electrons, or two photons counted.

You're welcome, but I can't really be sure that figure is correct. The source was not authoritative like DXO   .

In case anyone still misses my point, worrying about per pixel SNR and DR is nonsense without reference to pixel count, the PPI at which the image is displayed and to distance at which it is viewed, and such. People are getting "blinded by science" when trying to draw conclusions above IQ from DR measurements like those at DXo.

BJL,
I always assume that whenever these image quality issues arise, whether it be resolution, SNR, color sensitivity or tonal range, the comparisons only apply at close distances, similar to the distance you would read a book.

At greater distances, perhaps the average distance from which one might most frequently view a print hanging on one's wall, many of these qualitative differences become irrelevant.

However, it is a fact, whenever one is attracted by a print on the wall, and one one walks up close to see the detail, one can be disappointed if one sees fuzziness and noise, because that's not 'real', and it's not pretty.

(Perhaps I should rephrase that for the benefit of those who like a clear distinction between the particular and the general. "However, it is a fact, whenever I am attracted by a print on a wall, and I walk up close to to see the detail, I can be disappointed if I see fuzziness and noise.")

I recall when I first encountered the DXOMark website, I overlooked the 'print' option, top/left of the graphs, but I now mostly do my comparisons in that mode.

Initially I was puzzled that the normalised print size used was so small, 8"x12" at 300ppi representing an 8mp image. But DXO explain on their website that this small print size was chosen in order not to exclude too many cameras which might need their output interpolated to reach a larger normalised print size.

Interpolating an image file can raise all sorts of objections and questions relating to the quality of the interpolation algorithm used. As it is, DXOMark results seem to be disbelieved by many who are under the impression that their preferred RAW converter can make the DXO results meaningless. But I find it strange that such people are unable or unwilling to demonstrate this impression with comparison images.

The obvious explanation is that such people are just kidding themselves. Another explanation is that such people are simply not competent to make such comparison images, and do not wish to reveal that fact, or expose their methodology to criticism.

What I gather from DXO's explanation of the 8"x12" normalised print size, is that any print size will deliver the same comparative result (on the graphs) as long as no interpolation is involved. In other words, A P65+ image and a D3X image, both downsampled to 8"x12" size, will have the same relative qualities as described on the graphs in 'print' mode, as a P65+ image downsampled to A2 size, compared with the D3X image at A2 size, because one can make a print from a D3X image at A2 size without interpolation.

[Ray]

Are you really sure?
Did you print already an A2 side by side?

We will throw away all our stuff to get this miracle!

Many thanks for your information.

Sorry you don't appear to understand that dynamic range is only one attribute of camera performance.

If your purpose is to make A2 size prints which stand up to critical viewing from close up, then a camera with a higher pixel count than the D7000 might be a better choice, such as the Nikon D3X, A900, 5D2, or an MFDB.

In the old days of film, experienced photographers knew that positives, or slide film, had a rather limited DR capability, but often slide film was preferred for various reasons, both esthetic and practical.

However, when the subject has a very high brightness range, as in the example at the end of this post, slide film would clearly not be suitable. Even the Canon 50D was not ideal for this shot because it has insufficient DR capability.

Generally, slide film had a useable DR of 5-6 stops, color negative film about 7-9 stops and B&W film about 9-11 stops.

We now seem to have reached a stage of technological development whereby a fairly modest, lightweight camera such as the D7000 can deliver approximately the same degree of useable dynamic range as a good B&W negative, but has the advantage of color.

Now it's clear that most of the time, for most shots, the dynamic range of any modern DSLR is sufficient. But sometimes it isn't, and on such occasions it can be infuriating having to mess around on the computer with noise-reduction programs trying to make a silk purse out of a pig's ear.

It's also clear that in certain circumstances, if you know what you are doing, you can compensate for camera inadequacy by using techniques such as 'merge to HDR', when the subject has a high brightness range, or stitching multiple shots to achieve a higher resolution image, if you want to make a large print.

However, such techniques usually require the subject to be still. If you or the subject is moving, then such techniques won't work.

Such was the case with the following shot of St Isaac's cathedral in St Petersburg. Not only was the subject moving (the young ladies in the foreground), but I was also moving fairly rapidly whilst sat in a tour coach which was turning around as it left the square. I got a brief glimpse of the scene and quickly raised my 50D to the window and took 3 bracketed shots.

Because of the significant movement between each shot, there's no possibility of successfully merging the 3 shots to HDR.

The best I can do is select the foreground in the shot with the longest exposure, and paste it onto the shot with the shortest exposure, then fix any problems where the two layers overlap. Even then, the foreground is a bit noisy and I've had to apply some noise reduction.

The difference between the longest and shortest exposure in this example is 2.67 EV. The crops demonstrate the quality differences.

The difference between the DR capability of the Nikon D7000 and the Canon 50D is 2.46 EV (according to DXO).

Got it?

[ErikKaffehr]

Ray,

Thanks for providing a good example.

Just a few points. The large dynamic range of negative films comes from saturation behavior in the highlight region. Digital pretty much clips at a certain light intensity, whereas film tends to have a pretty pronounced shoulder. Density still increases with light but much slower. This is essentially a characteristic of digital imaging.

What BJL and I say is that recent wins in the dynamic range may be more related to reduction in readout noise. But at low exposures shot noise will dominate. Shot noise is simple photon statistics, nothing chip designers can do anything about.

Best regards
Erik

Sorry you don't appear to understand that dynamic range is only one attribute of camera performance.

If your purpose is to make A2 size prints which stand up to critical viewing from close up, then a camera with a higher pixel count than the D7000 might be a better choice, such as the Nikon D3X, A900, 5D2, or an MFDB.

In the old days of film, experienced photographers knew that positives, or slide film, had a rather limited DR capability, but often slide film was preferred for various reasons, both esthetic and practical.

However, when the subject has a very high brightness range, as in the example at the end of this post, slide film would clearly not be suitable. Even the Canon 50D was not ideal for this shot because it has insufficient DR capability.

Generally, slide film had a useable DR of 5-6 stops, color negative film about 7-9 stops and B&W film about 9-11 stops.

We now seem to have reached a stage of technological development whereby a fairly modest, lightweight camera such as the D7000 can deliver approximately the same degree of useable dynamic range as a good B&W negative, but has the advantage of color.

Now it's clear that most of the time, for most shots, the dynamic range of any modern DSLR is sufficient. But sometimes it isn't, and on such occasions it can be infuriating having to mess around on the computer with noise-reduction programs trying to make a silk purse out of a pig's ear.

It's also clear that in certain circumstances, if you know what you are doing, you can compensate for camera inadequacy by using techniques such as 'merge to HDR', when the subject has a high brightness range, or stitching multiple shots to achieve a higher resolution image, if you want to make a large print.

However, such techniques usually require the subject to be still. If you or the subject is moving, then such techniques won't work.

Such was the case with the following shot of St Isaac's cathedral in St Petersburg. Not only was the subject moving (the young ladies in the foreground), but I was also moving fairly rapidly whilst sat in a tour coach which was turning around as it left the square. I got a brief glimpse of the scene and quickly raised my 50D to the window and took 3 bracketed shots.

Because of the significant movement between each shot, there's no possibility of successfully merging the 3 shots to HDR.

The best I can do is select the foreground in the shot with the longest exposure, and paste it onto the shot with the shortest exposure, then fix any problems where the two layers overlap. Even then, the foreground is a bit noisy and I've had to apply some noise reduction.

The difference between the longest and shortest exposure in this example is 2.67 EV. The crops demonstrate the quality differences.

The difference between the DR capability of the Nikon D7000 and the Canon 50D is 2.46 EV (according to DXO).

Got it?