Bit Depth and Dynamic Range
Submitted by mitch on Thu, 07/30/2009 - 09:43.
David Saffir has written a blog entry about bit depth and posted it on the HP Professional Photography site.
David titled his blog entry, Bit Depth Basics: More Than A Numbers Game.
Well, yes!? It is more than a numbers game.
I don't know why David chose to talk about dynamic range in the same blog entry. Please, read his blog and tell me if I've unfairly interpreted his comments and his example. He doesn't come right out and say it, but you sure get the impression that more bit depth means more dynamic range.
"The major benefit of working with high-bit images is increased dynamic range— the range of tones and detail that the camera can record from the darkest dark to lightest light."
I've seen this sort of comment floating around on forums. I'm not sure where the logic comes from, except that HDR photos are 32-bits as opposed to 16-bits.
The relationship between bit depth and dynamic range is not 1:1. More bit depth does not mean that a 14-bit Nikon D3 can capture more dynamic range than a 12-bit Nikon D2X. Yes, theoretically, the D3 can. Theoretically, the maximum difference between the lightest and darkest pixels can be 16,384 for the D3 and only 4,096 for the Nikon D2X.
As we say in multivariate statistics, ceterus paribus, which is a fancy Latin expression that means assuming everything else is equal. It also means assuming both cameras capture light perfectly and convert that to digital signals perfectly. Well, none of that is even close to real-world conditions.
The bit depth of a RAW file from a digital SLR tells us nothing of practical note about the dynamic range captured in that RAW file.
There are definite editing virtues to greater bit depth. Photos with more bit depth are less likely to posterize during editing. The difference between 16,384 distinct levels and 4,096 distinct levels means that you can likely make some more extreme editing transformations without noticing visible posterization.
I said likely because much of that difference is more marketing claim than visible difference. People will show you histograms and numeric analyses to demonstrate how 14-bits is superior to 12-bits. Here's a point where I agree with Jeff Schewe and Bruce Fraser. As they write in Real World Camera RAW with Photoshop CS4, this is largely a marketing ploy.
I don't strive for pretty histograms. I aim for pretty prints. I'll acknowledge, there might be the rare RAW file where the edits a digital photographer might apply to that same image from the same camera to make a print or image for the Web would show visible posterization in 12-bits and none in 14-bits. Honestly, I haven't seen even one yet. Not where someone didn't do an extreme close-up or I'd need a loupe to see the posterization difference on a print.
If I was to sell my Canon 1Ds MkII to buy a 1Ds MkIII, it wouldn't be for 14-bit depth. That's not worth the thousands of dollars to me. Even the extra resolution is marginal in its visible impact. No, what would make me consider the trade-in would be the improved noise reduction. Each generation of DIGIC just gets better!
Bit-depth is also not a good indicator of the dynamic range of the scene reproduced in an HDR file. A 32-bit HDR image has to be tonemapped before it can be displayed on a monitor or printed. You can take a series of photos in 8-bits or in 12-bits or in 14-bits and once the resulting 32-bit HDR photo is properly tonemapped, even the 8-bit version will show the original dynamic range that was captured.
For more on HDR, bit depth, and dynamic range, I recommend readers view the following blog entry:
http://hdrphotos.info/dynamic-range-and-bit-depth/
The discussion of bit-depth and dynamic range also misses an important element. It ignores the effects of noise on what engineers refer to as quantization error. That's rounding error that results when the in-camera noise reduction takes tiny fluctutions in nearby pixels and rounds them all to the same digitized value in the RAW file. The posterization difference between 12-bits and 14-bits are likely to be less than residual noise, and that means it will almost certainly not be visible to the eye. In other words, the dithering effect of noise will be stronger than any likely difference in perceptible posterization between 14-bits and 12-bits.
For more on this point, I recommend readers see Emil Martinec's technical article on noise.
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/
As Emil notes, "Curiously, all the 14-bit cameras on the market (as of this writing) do not merit 14-bit recording. The noise is more than four levels in 14-bit units on all of these cameras (Nikon D3/D300, Canon 1D3/1Ds3 and 40D); the additional two bits are randomly fluctuating, since the levels are randomly fluctuating by +/- four levels or more. Twelve bits are perfectly adequate to record the image data without any loss of image quality, for any of these cameras (though the D3 comes quite close to warranting a 13th bit)."
He argues that the recording method in the Nikon D300 explains that the difference in D300 files compared with, say, the Nikon D2X, is not the result 14-bits per channel. It's the result of the D300 reading the sensor data more slowly. Reading the data slower by a factor of 3 or 4 means the data can be read more accurately and when it is read more accurately, you get less noise. That will have a bigger impact on perceived image quality between the Nikon D300 and the Nikon D2X than 14-bit depth v. 12-bit. Remember that Latin phrase, ceterus paribus. It's not just bit depth that distinguishes what information is captured in a Nikon .NEF file.
Bit depth and dynamic range are not completely unrelated. Regrettably, David Saffir's recent blog entry promotes the blarney from camera manufacturers that a DSLR with 14-bit depth is going to result in perceptibly better photos than one with 12-bits. That just ain't so.
David's other observations are good advice. You want to start with as much detail as possible and you want to work with your photographic images in a way that preserves as much detail as possible (usually -- sometimes we reduce detail on purpose through blurs, blends, and other edits).
Technorati Tags: - mitch's blog
- Login or register to post comments
sensor bit and Dynamic Range
Michael Freeman in his book Mastering HDR Photography comment on this on page 16:
To clear up any confusion about dynamic range and bit depth, the number of bits determines the precision of the capture - how many steps the range of brightness can be recorded as. In other words, a 14 bit sensor is more accurate than a 12-bit sensor. Tough it might, it does not necessarily capture a wider range , but more detail within the range.
Re. sensor bit
Thanks for the comment. :) It's helpful.
Jeff Schewe and Bruce Fraser use the metaphor of a staircase. Dynamic range determines how high and how low the staircase reaches. Bit depth determine how large or how small for the individual steps.
The problem with these sorts of metaphors is they miss the important point that when the individual differences get smaller than a recording device can record, smaller than the average level of noise, or smaller than the eye can perceive, then there will be no perceptible improvement from the increased bit depth.
There can be diminishing returns as bit depth increases. Going from 8-bits per channel to 12-bits per channel can sometimes result in less perceptible posterization. I've seen this myself. (It's easy to overstate this, and that's the reason for the italics.) I won't say that going from 12-bits to 14-bits will have no perceptible effect at all. Ever. For any image file. It can. But it requires the convergence of many factors and is extremely unlikely for the typical adjustments to tone and color that a photographer makes.
This is an interesting twist for me. I've argued for years against Dan Margulis' argument that 16-bit editing was unnecessary (at a time when 16-bit RAW files were universally 12-bits). Now, I find myself in the position of arguing that 14-bits is unlikely to provide a perceptible improvement over 12-bits. That's diminishing returns!
Another thing that confuse
Another thing that confuse people is the color space. Since we're talking about "data", when you mix the 8, 18 or such in color space... then everything get even more difficult to understand.
Now, regarding your comment between 12 and 14 bits, you are right, it's a very small step up, but sometimes is good to have that information, mostly when the retouching involves drastic changes. For example Using Topaz Adjust or Lucis or any other plug- in like that, the benefit of just that "tad" of information could avoid banding in really strong settings (for people who likes that look).
Re. Another thing that confuse
You're right, Jorge. Color space is important. If you edit in an extremely wide color space like ProPhotoRGB, you'll be more likely to see a perceptible difference with 14-bits v. 12-bits. Each level is a larger change.
I avoid ProPhotoRGB as a working color space. I do use it as an archival color space.
My reason for that is to avoid highly saturated colors causing clipping or clumping in AdobeRGB.
What I mean by an archival color space is this: when I make the RAW adjustments in ACR, I have it use ProPhotoRGB. Lightroom uses its own color space, and that's related to ProPhotoRGB. Once my RAW photos are in Photoshop, I convert them to AdobeRGB. Hence, that's my working color space.
Doing this will preserve the maximum number of highly saturated colors (the Adobe Color Engine will take colors that are out-of-gamut for AdobeRGB and convert them) so it will reduce the number of out-of-gamut colors I have to wrestle with when I get ready to print AND it will reduce the possibility of visible posterization.
If you want to reduce posterization, you want a narrower color space, not a broader color space. That way, when you chop up the color space into tiny, discrete increments, those discrete levels appear continuous.
If you want lots of highly saturated colors, you want a wider color space. The downside is that you make it more likely that some of those discrete levels will no longer appear to be continuous. You also compound the number of out-of-gamut colors when you get ready to send the image file to an output device, like a printer.
The point of Emil's whitepaper is that you don't really get more information with 14-bits than you do with 12-bits. Yes, in theory, you certainly do. But that ignores what happens inside the camera when the sensor data is read and processed.
Most of the DSLRs with 14-bits are high end with high resolution sensors. That means they have smaller photosites. To achieve such low noise, there have been improvements in the sensors, but the cameras also apply very sophisticated noise reduction in-camera. That's typically applied to RAW files as well as JPEGs. As a consequence, that additional information with 14-bit depth will very likely be treated as noise rather thn information.
Emil argues that the D3 *ALMOST* qualifies for a 13th bit. IOW, the benefit from 12-bit is largely illusory. One might argue -- and I have argued this re. 12-bits v. 8-bits -- that it's best to keep all that information, "just in case." The downside is larger files and slower processing.
Until I see the first conclusive demonstration (RAW files, 14-bit capture and 12-bit capture from the same camera, same subject, same lighting, same settings except for bit depth) that there is a perceptible difference from 14-bits, I'm going to remain a skeptic. ;) I can be persuaded, but I'll need to see it with my eyes on a typical photographic subject. :)
The HDR argument doesn't convince me. If you're using several photos with small changes in ISO, such as 1/3 to 1/2 stop, you'll have maximum information for the scene to tone map, whether you use 8-bit, 12-bit, or 14-bit depth. Let's keep this discussion going, Jorge. It should be interesting, at least for a few people who care about things like bit depth and color spaces. :)
Re. Another thing that confuse
Ah... I'm learning here too!
Ok, I agree on everything you say but I have to go into "deeper" situations.
Your explanation and workflow is interesting, I try to stay in aRGB all the time, then to save, I may save a TIFF for my special pictures and jpg to the conventional ones. But now, it comes to mind something else: society and business.
We see newer stuff and software everyday. Manufacturers are always claiming developments, sometimes the products are better sometimes not but the trend always continue. In our subject, for the cameras, each year we see a new model, most likely packed with more Mp and more bits and many other gimmicks. For the rich gadget person it is a must to have the latest, for the rest of us not so, the prove is the Mp war and this discussion. 12 and 14 bit are not that different. But what about 12 and 16? Over there I see a benefit. My point is, business are moving forward, right now, when we work on 16 bit and later convert to 8 may be the wrong approach, we may stay in 16 because clearly the next photoshop will be able to work on even 32 bit images and 16 will use all filters.
We are in a "transition" moment, in the past I kept my Mavica pictures as jpg in 8 bit, there was nothing else in that time, today I can chose between 8 and 16. What is better?
Clearly, all this statements are to educate people, we do not need to jump for the newer camera which offers a jump from 12 to 14 bits. This "not in depth" discussions are the best ones since most people can understand it, no "rocket science formulas" are used.
So moving on regarding color. What we do and what we see in photoshop is not always the best for printing. Just a week ago, I got a matte paper which do not hold even the gamut of sRGB, I have to work extra in the gamut preview with the paper profile to get a decent print. So, it is imperative that people clearly see what they really want of their images. I've seen people with the 5D Mark II shooting only jpg and printing 6x4 pictures (nothing wrong with the concept) but any other camera can do it a Rebel would be better since they wouldn't need to carry a bigger camera. Then in the post processing, many people don't understand this bit, color or dpi concepts and because they read, they sometimes pick wrong numbers, I saw someone saving his images at triple size and converted to 600 dpi and, of course, complaining that the camera didn't give the detail the manufacturer claim. And at the end, when more knowledgeable people saying a printer or paper is not good.
I've tested a 30D and a 40D side by side. The 40D is marginally better in high ISO with good lighting (to increase shutter speed) but way better in low light condition, that's about it. File size and depth are the same to my eyes. Now, when post processing goes, I did see a slightly difference when applying my workflow but it was noticeable at huge zooms, nothing that you could even see in a 48"x36" print.
Now, for HDR... agreed in no difference between a bracketed shot between 12 and 14 bit cameras... but 8 bit vs. 14 bit makes a difference, mostly at detail level; it is arguable because unless in lab, a scene does not remain static in real life. Even taking a picture the next day at the same hour can render different outputs. For the sake of the discussion, I would support that 14 bit is better than 8 for HDR since tone-mapping is a strong "pixel-pushing" process, the more information the better for that purpose. A proof of that is that sometimes even using ISO 100 pictures in good light, after the process we can see noise, Tone-mapping boost the fine grains we didn't see in a "clean image" (which lead us to the realm that ALL pictures require noise cleaning).
The problem here is that we do not have real information but only guesses. We cannot assure that a RAW file is cooked or not (when not special settings are used as noise reduction, high tone priority, etc). Some people claim that some manufacturers do something... but since the manufacturers are not talking, I think it's almost impossible to prove something. So, going with the flow and thinking that 14 bit and more Mp is the holy grail I have to invite people to see the differences. When you have less detail and less Mp photography is more forgiven, since there are not much detail, a picture may "look better" than a very slightly blurred picture taken with a higher bit & Mp camera as my experience between a 30D and the 5D.
So Mitch, in the argument of HDR... What is our final output? If we're sending the image via e-mail to our friends or posting online... yes, there's no difference between 8, 12 and 14 bits. If we're creating a fine print larger than 19"x13" (professional home printer) I would take out the 8 bit camera.
Re. Another thing that confuse
There will likely be a significant difference with 8-bits per channel unless you use small steps in exposures.
At 8-bits per channel, you are chopping up each channel into only 256 discrete levels. 12-bits gives you 4,096 discrete values. That's quite a jump. 14-bits gives you 16,384 values for each channel. Less of an improvement over 12-bits than over 8-bits.
If I had a choice of cameras, I would lean towards greater bit-depth, all else being equal. I do not, however, feel that the 12-bits per channel for my 1Ds MkII is inferior to the 14-bits per channel for the 1Ds MkIII.
Yes, if I had a few thousand dollars laying around, I'd like a 1Ds MkIII. But, as I said, that's largely for the improved noise reduction *AND* the 1Ds MkII is a great camera, even when it comes to noise.
Agree with you Mitch
Hi Mitch, I tried to post this over at David's blog but couldn't get it to take:
Hi David and Mitch,
I'm afraid I tend to lean in Mitch's direction here. Whilst I agree that there are 'theoretical' benefits from higher bit rates, in real world display situations the differences are mostly imperceptible.
Mitch is also perfectly correct - 8-10-12-14-16 bit has absolutely nothing to do with the dynamic range of a device, just the 'sensitivity' - i.e. steps it can record.
A histogram BTW is a very poor tool to use, firstly histograms in any software AFAIK (e.g PS), only has 256 steps (see monitor comment below), thus when you show a 'histogram' of a 16 bit edited file you are unlikely to see any "gaps", however if you were able to expand the histogram, or say zoom in to a section you would see the same type of gaps.
Also, whilst the human eye has an upper dynamic range (DR) of 25 to 30 stops, many many times that of a 12 stop camera this does not translate to the same capability in contrast detection i.e. perception of steps of brightness, at best the eye has about a 10,000 step range (more sensitive at lower light levels than high) see http://www.clarkvision.com/imagedetail/eye-resolution.html.
More importantly, (LCD) monitors typically have a DR of only 8 stops, they can only display 256 levels in R, G, or B and (most) are only 8 bit devices so.... guess what? They can only ever display 256 gradations. So if the monitor can't display the DR or the subtle gradations, just exactly how are we going to see it??
Paper: Even the best quality paper (being only a reflective device) can manage between 3-4 'stops', and whilst the 'theoretical' gradations the best paper/printer combo is higher, it is nowhere near 4000 levels, anyway the very limited DR makes comparison from 8 to 12 -14- 16 bit sources irrelevant.
Again if the displaying mechanism can't actually display the difference, how is it possible to see it?
The only real test IMHO, is to perform (like in medical science) a 'double blind" test, i.e. you are given multiple versions of the same image , say 10 x 8bit, 10 x 16 bit, 10 x 14 bit (not just 2 images, so you have a 50/50 chance of guessing right). If you can't correctly identify which is which, then poof goes your argument.
However I must concede that the creation of computer generated gradients often benefits from a higher bit rate, but that isn't a photo now is it?
To repeat, unless a true double blind test is performed. this is all just "I think I can see a visible difference, therefore there is a visible difference", well how can I dispute that? no I say - how can you prove it? Easy, a proper controlled test.
It's a little like hi-fi / audiophile claims, when the 'experts' are put in a controlled room, magically they can't determine the difference they claimed they heard between two similar components.
Brian - Melbourne Australia
Re. Another thing that confuse
Again, despite I agree on the argument, a difference IS a difference as minor as it could be.
Phelps won for fractions of seconds… but he won. So, perhaps 14 is a minimum benefit, but nevertheless a benefit. We may not see it now, but perhaps we’ll se it in future versions of software and hardware.
We don’t need exhaustive tests to prove the benefits of having more information, it’s as simple as working in ACR (or any other RAW converter). Open any jpg file and whatever is there is what we have. In any RAW file, we can “recover” highlights or shadows and the second proof is even simpler, we cannot have the cake and eat it too, when we recover something we are giving up something else, we recover highlights and shadows and there goes our mid-tones. No monitor (this time Mitch, you’ll need more than a couple thousands in order to get a Brightside HDR monitor), paper or camera will give us a better DR.
Now, if we are staying in the current technology, I would agree, we don’t need more, but with the steps taken, soon we will have better DR in camera and better monitors and better paper.
Up to now, I have to say that there’s no real studies regarding this subject ar they are kept in secrecy; authors give different numbers for same situations: the other poster mentioned human eye between 25 and 30 while in my books it’s mentioned only 20 (with long term adaptation), his info regarding monitor is 9 while mine is 8 and his paper is between 3 and 4 while mine is between 6.6 and 7.6.
For sure I can say the big difference between printing in the best glossy paper made by Office depot, Hewlett Packard, Moab, Ilford and Hahnemuhle… I’m not talking about the quality or feel, but when doing the profiling using the same ink, and doing the proof in Photoshop you can see the difference!
Right now, if I had those 2 grand to spend, I’ll get me a tilt and shift lens and I wouldn’t change my cameras; as I said before, the difference between 12 and 14 is no significant but in a couple of years I may doubt between 12 and 16.
HDR software, in the other hand, is something subjective which can only go up to a limit; no machine can decide as a human brain; there are many things considered by the brain and personal taste is one.
We now by now how the eye works and is tied to the brain making up for the “processing” of images. No software will be able to do that, HDR software is getting better in deciding to take more information for one or other picture for blending but it’s fat to decide how a person view the scene… heck! I don’t even know if two people seeing the same scene at the same time both see the same colors!
A couple of clarifications - and a simple question
Hi jfdp1686,
My real point was to ask how can a difference be seen if the display device does not have the capability to display the difference? Can someone please explain that to me?
Re the eye DR, I'm using 'theoretical' maximum/extremes, as that is what is being used for the 12-14 bit devices, I agree with your 20 and even lower if a 'static' DR is assumed,i.e. like a snapshot, however that is not how we see.
Re monitors, my original post said 9, a typo I meant 8
Re paper, some manufacturers 'claim' 6-7 stops, lets do some #'s, as paper is only a reflective device, its DR is a calc on its max and min % of light it can reflect, if it still reflects 5% of light in it's darkest black, and the bare white paper 80% than the the ratio is 80/5 = 16 which is 2 to the power of 4, i.e. 4 stops.
I'm not aware of any paper that reflects (sometimes called the gloss factor) more than 90% and less than 5% with maximum black ink, this then = 90/5 = 1:18 only a tad over 4 stops, it would need to be 99%/3% to get to 1:32 = 5 stops, even a mirror barely reflects 99%.
OK lets forget all that, here's my simple question, we all agree that monitors have a bigger DR than the best paper so lets stick with them.
The original discussion was about 12-14 bit, which does nothing for DR, but does allow finer steps within that range. OK
Can someone tell me a monitor that has more than 256 brightness levels for its pixels? Even the ultimates - Eizo and Lacie don't claim to have more than this - here's a quote from Eizo themselves, "EIZO carefully measures and sets every shade of R, G, and B from 0 – 255 on the production line...."
OK so if the monitor's pixel can only show for example a change from step/level 124 to 125, just exactly how can it display (and you see) the difference between step 1986 and 1990 (which falls between 124-125 in a 4096/12 bit range). It can't, it can only display at level 124 or 125. The 1990 = approx step 124.4, which is impossible for the monitor to do so it will display at 124, hence 1986 and 1990 will both display at the same brightness level.
So please , please, someone explain how anyone can see a difference that the the monitor can't display?
Brian - Melbourne Australia
Re: A couple of clarifications - and a simple question
Brian,
My point saying that there are no real numbers is because different authors have different numbers (average) but never the same. Why is that? I mean, today gravity is the same for everybody… why not reflectance?
Regarding the paper, I need to research, but I think it is an special (and very expensive) Kodak paper that can reach the 6 while a metallic paper could give 5 and change.
Now, regarding the main subject, as I said, I totally agree that right now, the difference between 12 and 14 is nothing; better said, is a difference which not show… yet.
Technology as you and I can afford is not in pair with the technology Adobe is using. Adobe is using 16 bit and you can even get your HDR file in 32 bit with very limited filters but rest assured that in future versions you will be able to work on those files.
What monitor will you use for that? The technology used right now in the BrightSide DR37-P HDR display which costed US $50,000 in 2006 so perhaps is now a little bit cheaper.
http://www.bit-tech.net/hardware/2005/10/04/brightside_hdr_edr/8
http://www.oled-display.info/dr37-p-hdr-display-brightside
So answering your question, Right now, there’s no way you can see it in a single view. But as I said in my past post, when you’re working in ACR, you can “take advantage” of the higher DR (hence PART of the 14 bit image) when you can “recover” highlights or shadows. In real life, you’re not “recovering” anything since the file has the information, you are only “showing it” (losing something else) in a device not capable of show the entire information. But again, technology is improving and there will be a time when the monitors as the ones I pointed will be the standard.
Let’s be honest here. Since ACR I’ve been impressed on some of my old pictures, what I can do with them now, surpass anything I did years ago and that is only thanks to the old RAW files. Imagine what you’ll be doing in the future with the files we have today!
WOW! My head is spinning!!
The most compelling thoughts that I take from all of this are two: The reinforcement of my long-standing belief that technology leap-frogs technology. I am revisiting four and five year old RAW images taken with a then state-of-the-art Canon D60, in preparation for an exhibition and amazed at what today's software will allow me to do with them. The data was (were) there just waiting for the software to catch up! How much better would they really be with my Canon 5D? How much better will my 5D images be with the software of some years from now?
And secondly, the joy that what we have here is a place where dissenting opinions do not result in dissention; where true professionals can disagree without being disagreeable. I will have to re-read the thread again and again before this old brain even thinks that it comprehends!!!
Thanks, folks!
Lauren