Dynamic Ranges of DSLR Cameras...

[rpcrowe]

It seems to me that the inherent dynamic range capability of a DSLR camera is an important facet yet, dynamic range is seldom, if ever, mentioned either by the various camera manufacturers in their ads or by camera reviewers. It seems that gazillion pixels and mega-mega ISO capability are what really sell cameras and dynamic range is disregarded. Is it as important as I think it might be?

I have attempted to research the dynamic range capabilities of various Canon cameras and have come up with these Digital Photography Review statistics for the 7D, 60D and 5Dii. However, the dynamic range has been reviewed using JPEG. I never shoot JPEG and I wonder if the dynamic JPEG range can be correlated to the inherent dynamic capabilities of these cameras when shooting RAW. In other words, is the dynamic range capability dependent on the camera and sensor or is it more concerned with the individual camera's conversion to JPEG...

Canon 5Dii
http://www.dpreview.com/reviews/cano...kii/page25.asp

Canon 60D
http://www.dpreview.com/reviews/canoneos60d/page13.asp

Canon 7D
http://www.dpreview.com/reviews/canoneos7d/page17.asp

Canon T2i
http://www.dpreview.com/reviews/canoneos550d/page16.asp

This review compares various cameras from different manufacturers such as: Sony A550, Samsung NX10, Nikon D5000, Canon 7D, and Canon T2i.

http://www.digitalcamerainfo.com/con...Resolution.htm

The 7D seems to come out the leader with almost 2 stops over the Samsung NX10 a half stop over the Sony, A550, roughly a half stop over the Nikon D5000 and roughly 3/4 stops over the T2i. Of course, we are comparing apples to oranges when matching an expensive 7D against a less expensive T2i. I am not sure of the validity of comparing the 7D against the cameras from the other manufacturers because of price differentials but, if a person just wanted to see which camera has the better dynamic range, it would be valid.

I was always under the impression that the 5D and 1D series cameras had the best dynamic range capabilities. Am I correct in this belief?

Please help me with some input regarding dynamic range in RAW of the various cameras. I am most interested in Canon models since I am locked into that brand because of the amount of expensive glass I own. I am not going to change brands but would be interested in what you think about whether dynamic range in RAW is an important parameter of a camera.

[Steaphany]

Richard,

Your signature says you've been a photographer in excess of 50 years, so you're well aware, as am I, that film photography too had it's dynamic range. Do you remember the word Latitude and how every film data sheet featured that characteristic S curve of exposure to density.

In digital photography, we're dealing with digital storage of analog information, so when talking about dynamic range, a property that is just as important as latitude was to film, the units of measure are simply different.

This enters a realm where having a background in the field of electronics and data acquisition systems helps, since that is basically what any digital camera amounts to. The performance of each component does not matter as much as the resulting performance of the whole system, or more significantly, the weakest component of the system. The problem with judging a camera by JPG performance is, as we all know, JPG performance itself isn't very good.

If we completely disregard the loss effects of the JPG format, the fact that the format only provides 8 bits per channel means at the theoretical best, the dynamic range can be no more than 48 db. Yet the theoretical dynamic range of a camera with a 12 bit Analog to Digital Converter, ADC, is 72 db, while a 14 Bit ADC yields 84 db and a 16 Bit 96 db. A metaphor equivalent would been using a scale that measures in integer feet or meters, great when needing precise work in the scale of miles or kilometers, to measure the month to month growth of a young child and wondering why the child shows no increase in height from year to year while still out growing clothes. Evaluating the dynamic range of cameras by reviewing JPG results is simply using the wrong scale.

To complicate matters, noise, which we all fight no matter what brand camera or ISO we use, decreases dynamic range, so those theoretical dynamic ranges need to come down a bit. If we were lucky enough to have noise not exceeding 1 Least Significant Bit, LSB, JPG would then top out at 42 db and cameras with 12 bit, 14 bit, and 16 bit ADCs would yield 66 db, 78 db, and 90 db respectively. Do keep in mind that db is based upon the log of a ratio, so even though that 1 LSB causes only a 6 db drop regardless of the number of bits, a 1 LSB noise level in an 8 bit system is still 16 times greater than a 1 LSB noise level in a 12 bit system.

Since noise can so adversely effect dynamic range, you also have to pay attention to where the noise comes from. With a digital camera, I'm sure we have all put blame upon that poor imaging chip. Yet, if you think about it, where should the magnitude of imager noise be ? The imager is accumulating charge from individual photons, in the millions, and apart from the loss due to quantum efficiency never equaling unity, two photo sites next to each other, receiving the same spectral distribution and light intensity should read out very similar signals, yet we'll often see significant pixel to pixel differences when we view pixel scale magnifications. The fact is those ADCs, the single component responsible for making digital photography digital, are also sources of thermal noise, reference noise, clock jitter noise, in addition to quantization noise. Plus, it's the number of bits in the ADC which determine the minimum quantization of pixel sensed light level to adjacent pixel sensed light level. So far, I've only touched upon two components, we also need to consider any analog circuitry within the imager, between the imager and ADC, and the digital processing capabilities, software algorithms, and the output file format that follows the ADC.

Obviously, Raw would always out perform JPG.

In the end, if you want a true determination of a cameras dynamic range, it should be evaluated as a system and scaled similarly to the measure of latitude in film, plotted as a graph of showing the function of light intensity falling on the imager chip to the digital value that gets stored within the resulting output file.

[Colin Southern]

Hi Richard,

Just a couple of quick points ...

1. You can get the dynamic range and many other parameters for just about every digital camera ever made - and compare them - all from one database at DxO Mark.

2. I don't believe that dynamic range is that big of a deal these days because most modern DSLR cameras are capturing around 11 to 12 stops at base ISO, and most prints are only using 4 stops, and most monitors 6 stops (that's all they can physicaly reproduce). I might add that those 4 & 6 stops are the image that we're presenting AFTER any dynamic range compression of the original 11 or 12 stops that were captured (ie the 6 stops that you're viewing may be from an 8 stop range you captured, after using things like fill light slider to reveal shadow detail), but by and large, there is a sufficient "safety margin" for most scenes.

So how many stops of DR do we really need to capture? A bride in a white dress standing next to a groom in a black suit will only have a 4 stop difference between them (I use this as a good example of the maximum dynamic range of a purely reflective scene) - so if you add a couple more stops to allow for exposure error and perhaps some shadow detail we might like to reveal, and we come up with 6 stops probably being adequate ("ball park" figures here). If the scene contains active light sources (ie not purely reflective) then what's required could be anything, but typically it's not extreme, and often an additional 3 stops will cover it (and if it doesn't then that's what fill flash is for). Where we REALLY start to push the limits is in landscape where we're shooting into the light (eg a sunset); this can exceed the DR of the camera - but - (a) we can then use HDR techniques like GND filters or multiple frames, and (b) even if we manage to capture - say - 20 stops, what can you do with it? It can't be displayed on any monitor / printed on paper without a lot of tonal range compression, and the greater the DR of the source, the harder it's going to be to get something looking good. I'm not saying that that can't be don't, but it's not a use so compelling that we need to change the design of every DSLR camera to accommodate it.

The bit that WOULD be useful in this respect is how dynamic range equates to high ISO performance; basically, dynamic range is what we get between the sensor saturation at one end, and the noise floor at the other. ISO steps work by changing the gain of the Analog to Digital converter, which also amplifies the noise - so for each "stop" of ISO you also lose 1 stop of DR; not a problem at base ISO, but at something like 3200 ISO on a 1DS3 we've essentially gone from a DR of around 12 down to 7 - and that can start to get "interesting". (on a side note, you'll generally find that the best high ISO cameras are the ones with the better base dynamic range, but all is not necessarily as it seems because a lot of them are also using more agressive noise reduction to effectively "lower the noise floor" - squeezing out more DR and thus more high ISO performance; problem is, the noise reduction only applies to in-camera JPEGs, which is why folks buy these wonderfully performing high-ISO cameras - shoot RAW - and then start complaining that they're getting a LOT more noise than the photos that the manufacturers are using to promote the product (and start crying "foul!).

I talk too much!

[jiro]

Maybe, just maybe, these links could be of help:

http://www.sphoto.com/techinfo/dslrs...slrsensors.htm

http://www.sphoto.com/techinfo/dslrs...lrsensors2.htm

http://www.sphoto.com/techinfo/dslrs...lrsensors3.htm

http://www.sphoto.com/techinfo/dslrs...lrsensors4.htm

[Steaphany]

Since my post used db and Colin used stops, I decided to do a little math and post some conversions, basically 1 stop difference in exposure, a ratio of 2 to 1, equates to 6 db. Now all you need to do is divide the db values of my post by 6 to get exposure stop values in terms of Colin's post.

A JPG with a 1 LSB noise level carries 7 stops and camera's employing 12 bit, 14 bit and 16 bit ADC's yield 11, 13, and 15 stop dynamic range images respectively. ( And people push things even further with HDR photography )

[Colin Southern]

Maybe, just maybe, these links could be of help:

http://www.sphoto.com/techinfo/dslrs...slrsensors.htm

http://www.sphoto.com/techinfo/dslrs...lrsensors2.htm

http://www.sphoto.com/techinfo/dslrs...lrsensors3.htm

http://www.sphoto.com/techinfo/dslrs...lrsensors4.htm

Me thinks this chap spends too much time over-analysing, and not enough time shooting "outside the lab"

[jiro]

Me thinks this chap spends too much time over-analysing, and not enough time shooting "outside the lab"

Hehehe. At least he provided pictures to back up his experiments. Personally I don't know much about Sensor Size relative to Dynamic Range. All I know is that the higlights goes to Zone VII or VIII and the lowest detail I need to retain on the shadow should fall on Zone III. After that calculation I just press the shutter to compose my shots.

[pwnage101]

I feel as though Colin hit the nail hard on the head. I can only suggest an interesting read:
http://www.clarkvision.com/articles/...ry1/index.html

The article is a little old (it uses the Canon 1D II for testing) but its objective nature is quite interesting. For the sake of this topic, you can skip to the "Dynamic range" section.

[Steaphany]

ISO steps work by changing the gain of the Analog to Digital converter, which also amplifies the noise

Is this actually what is done within the camera's electronics or is the selection of imager ISO a setting of the floor, Zero reference, value of the ADC ?

According to "The Handbook of Astronomical Image Processing" by Richard Berry and James Burnell, which has a chapter dedicated to "Counting Photons" addressing signal, noise, and dynamic range and another on characterization of camera performance, an imager accumulates charge with each detected photon until the photosite saturates. At this point, absorption of additional photons to not yield a corresponding increase in charge.

The key feature of digital camera ISO is that it mimics the property of film sensitivity, i.e. an increased ISO allows you to get an image with comparable dynamic range with less exposure, or fewer photons.

Electronically, this could be achieved with an increase in gain, as you would achieve with an amplifier or, since we are working with an ADC, you could raise or lower the Zero reference point of the ADC, effectively sliding the operating dynamic range of the ADC over the operating range of the imager.

So, I have this question and I hope we have a few digital camera design engineers lurking here with us, What is the actual method employed by digital cameras to set a camera's ISO operating characteristics ?

[Colin Southern]

Is this actually what is done within the camera's electronics or is the selection of imager ISO a setting of the floor, Zero reference, value of the ADC ?

Hi Steaphany,

That's an easy one to answer ... I wouldn't have a clue!

To be honest, in practical terms, all I really need to know is that as I increase my ISO, I have to consider my exposure and dynamic range more, to ensure I "get the shot".

I know a lot of people like to invest a lot of time and effort covering things like the noise performance of "brand X -v- brand Y", but in my opinion its usually a case of "is a Ferrari better than a Bugatti when all you want to do is go to the corner dairy"? In reality, things like dynamic range and megapixels really don't have any significant effect on the quality of my work ... and I'd like to think that I'm probably shooting a wider variety of things than the average photographer. Hence the comment above about people needing to "get out of the lab more"; sure, it's interesting stuff, but I don't think I'd sell many images of 100% crops of teddy bears 5 stops under-exposed ... I DID however sell 2 large canvases of a fishing boat shot at 3200 ISO the other day, despite the "noise" (which is all but undetectable in the real-world print, despite ZERO noise reduction being applied).

I general I'm reminded of the old saying "those who can, do ... the rest write reviews and post them on the internet!"

[revi]

I hear you all about the camera covering 12+ stops of dynamic range, screen ~6 and paper ~4 stops. I understand what you mean in terms of light intensity from white to black, and where the problems are in capturing a scene (full sunlight + deep shadow, would that be only 4 stops, Colin? )

As I only need 8 bits for acceptable/good display&print, RAW gives me a 4-6 bits margin to play with. Or, I can basically divide the RAW pixel values by 16-64 and end up with the full range downsampled to 8 bit range.

But I still use the same 8 bits per color to display & print my photos, with 0 for deepest black and 256 for highest intensity (white in grayscale).

So, where does the difference in dynamic range come in play here? In other words, what is the relation between dynamic range and bit depth? (leaving noise outside here, a good 100 ISO capture shouldn't have any after going to 8 bit).

Remco

[Colin Southern]

Hi Remco,

My suggestion is to not equate dynamic range with bit depth. In some ways it's a bit like the height of a staircase doesn't equate to the number of steps (not a great analogy, but hopefully you get the idea),

Basically, the "rules of engagement" are ...

1. Capture sufficient - clean - detail in the first place. If you're shooting a purely reflective scene then "close enough" will probably be "good enough" because you'll have a HUGE safety margin in a RAW capture. If you're shooting something with backlighting then you'll need to be more careful (with the amount of care required being proportional to the DR of the scene). You also need to be careful when shooting at high ISOs.

2. Process your capture as a 16 bit file, and do as much of your manipulation in ACR (or RAW converter of your chosing). Manipulations whilst still in linear gamma mean less image degradation.

3. Once into Photoshop "proper" - again, with a 16 bit file, do any additional manipulations.

At this point - and only this point - you can convert the image to 8 Bit and save it in any format you like.

Basically, a RAW capture usually has big safety margins, but some of that safety margin is lost when you pop out of the RAW converter (ie when gamma conversion is applied), and some more safety margin is lost when you convert from 16 bit to 8 bit ... so the "trick" is to only convert to 8 bit (if at all) once you don't need the safety margin anymore.

Does this help?

full sunlight + deep shadow, would that be only 4 stops, Colin?

Not sure what you mean to be honest. Put it this way ... if the bride in her white dress is standing next to her groom in a black suit - both illuminated by the sun coming at them from in front of them - then if you spot-meter both, you'll have a 4 stop difference. Does this mean that only 4 stops are required? Well that depends ... in reality you're not dealing with a truly black suit (if it were truly black you wouldn't be able to see it) - so you're really dealing with a shade of gray. If there is more detail that you wish to reveal using for example the fill light slider then obviously that may draw on additional dynamic rnage that (hopefully) you've captured cleanly. The thing to remember with a RAW capture is "just because you can't see the detail doesn't mean that it's not there"; it's often nothing short of amazing what can be pulled out.

If you're asking "would a scene with full sun and shadow detail only have a DR of 4 stops then no - of course not (I was only talking reflective scenes) - but 4 stops would be what you'd have to compress it in to if you were to print it. There's no way we'll ever be able to look at a print of the sun and have it burn our eyeballs (although some of the stuff I've read on facebook has come pretty close to burning my eyeballs!).

[rob marshall]

Surely the problem here is that while the human eye can cover 20 stops, the camera can only cover half of that? That's why shots rarely look like you actually saw them at the time of shooting. I've found though that post-processing, especially in RAW, can create the impression of a greater range than the camera took. You are limited anyway, as Colin said, by the output limitations of paper and screen. Personally, I don't see it as a problem. What's important is what the image looks like when viewed. Camera technology and software editing will improve in future years, but for now I think it's actually pretty good.

Perhaps some of you need to take Mr Lik's advice - "Get off the couch and get out there..."

[revi]

Thank you for the answer, Colin. I think I didn't quite get my question across...

Hi Remco,

My suggestion is to not equate dynamic range with bit depth. In some ways it's a bit like the height of a staircase doesn't equate to the number of steps (not a great analogy, but hopefully you get the idea),

I see (and understand) what you mean. I also see the bit depth used sometimes as equal to dynamic range...

(snipped process)..

I indeed try to capture as cleanly as possible and process in 16 bits.

Basically, a RAW capture usually has big safety margins, but some of that safety margin is lost when you pop out of the RAW converter (ie when gamma conversion is applied), and some more safety margin is lost when you convert from 16 bit to 8 bit ... so the "trick" is to only convert to 8 bit (if at all) once you don't need the safety margin anymore.

Does this help?

I does, thank you

(snipped an explanation about the 4-stops range)

If you're asking "would a scene with full sun and shadow detail only have a DR of 4 stops then no - of course not (I was only talking reflective scenes) - but 4 stops would be what you'd have to compress it in to if you were to print it. There's no way we'll ever be able to look at a print of the sun and have it burn our eyeballs (although some of the stuff I've read on facebook has come pretty close to burning my eyeballs!).

What I have problems understanding is the part I put in bold:
Once we have a capture with a dynamic range of say 12 stops, we have basically a range of values where 0 is black, and 2^12 is white (12-bit data, so striclty speaking white is (2^12 - 1)). We then transform it to an 8-bit range of values, so 0=black, and 2^8 is white. BUT those still represent the same black and white we had in the original scene, we just have less shades in between. So we still cover the same 12 stops we covered in the original scene.
Now we print those values. We still have 0 is black and 2^8 is white. Yes, the paper will not reflect 2^12 times as much light from white as from black, but only 2^4. So dynamic range got compressed, but WE didn't do anything to compress it...
(We still use the same number of steps, but changed the height of the staircase).

Unless we run in trouble because we cannot distinguish between adjacent steps anymore (esp. in shadows and highlights).

I hope I've made my question a bit clearer.

Remco

[rpcrowe]

Thanks ladies and gentlemen. I appreciate your extensive input. This site is the most personal and the most friendly of any photo oriented site I have ever visited. It is the first site I gravitate to when I open my computer in the morning and I am always pleased with the amount and content of the feedback.

[Colin Southern]

I hope I've made my question a bit clearer.

Hi Remco,

I see what you're getting at now.

Just keep in mind the difference between what's captured and what we can see. If we use a simple 0-255 scale to cover a 12 stop dynamic range, if 0 = black and 255 = white, then 0 would be black - and 10 would be black - and 50 would be black - and 100 would be black ... it's probably not until you got into the mid hundreds would you start to see anything (ignoring gamma curves). So not saying that the information isn't captured, but we just cant see it. So if we want to use it, we then have to up-shift what's in the region we can't see so that it's in the region we can see, at which point what covered a bigger range of values now coveres a small range of values ... so the range has been compressed.

Does that explain it well enough?

[revi]

Hi Remco,

I see what you're getting at now.

Just keep in mind the difference between what's captured and what we can see. If we use a simple 0-255 scale to cover a 12 stop dynamic range, if 0 = black and 255 = white, then 0 would be black - and 10 would be black - and 50 would be black - and 100 would be black ... it's probably not until you got into the mid hundreds would you start to see anything (ignoring gamma curves). So not saying that the information isn't captured, but we just cant see it. So if we want to use it, we then have to up-shift what's in the region we can't see so that it's in the region we can see, at which point what covered a bigger range of values now coveres a small range of values ... so the range has been compressed.

Does that explain it well enough?

That helps a lot, thank you.

Remco

[Nicola]

morning to all,
I quickly read the posts,
perhaps the following web site can help, I think is one of the best about DSLR reviews
(I put a page of the canon 60D review...)

http://www.imaging-resource.com/PROD...60DIMATEST.HTM

hope that helps

bye
Nicola