12-bit image files can store up to 68 billion different shades of color. 14-bit image files store up to 4 trillion shades. That’s an enormous difference, so shouldn’t we always choose 14-bit when shooting RAW? Here’s a landscape I snapped, then found out later I had shot it in 12-bit RAW. Better toss this one out, right?
Depending on which class you took at the University of Google, the human eye is only capable of distinguishing between 2.5 and 16.8 million different shades of colors. If this is the case then wouldn’t 12-bit be plenty? Even 8-bit JPEGs can render 16.8 million colors.
There are many upsides to shooting 12-bit instead of 14-bit. The files are smaller, hence your camera’s buffer doesn’t fill up as fast, allowing longer action sequences to be caught before buffering out. 12-bit files take up less space on your memory cards – great for if you are on vacation without the ability to download your images every night. You can save money because you don’t need to purchase as many gigs of storage. Likewise, 12-bit hogs less space on your drives at home and the same number of 12-bit files load faster than if they were 14-bit. Lastly some cameras, such as Nikon’s D7100 and D7200 achieve higher burst rates when shot in 12-bit than in 14-bit.
So if the human eye can’t discern the difference and 12-bit has so many advantages, why doesn’t everyone just shoot in 12-bit? For the same reason brides want a real diamond, not a cubic zirconia. Admiring an engagement ring from a normal viewing distance, few people can tell the difference, but give that ring to a lab technician with a refractometer and they can distinguish the two (Note to readers: I will cut off your shutter finger if you forward this to my fiancé).
When given the choice I’ve always shot 14-bit, because as an American I know bigger is better and besides it’s my constitutional right to fritter away as many redundant bytes as I please. I went to the internet (I sense trouble coming) to validate my feelings and found a lab test where someone shot a lens chart with a DX body at 4 stops underexposed then zoomed in to 200% and sure enough, you could see a difference. Then I checked another site where test shots showed no difference. This was getting confusing for my puny brain so I decided to field test 12-bit versus 14-bit to see if I could tell any difference. I started out with landscapes – if anyone is picky about file quality it’s us landscape geeks. Bear in mind that these tests are in the field, not the lab, and though I tried my best to keep all parameters the same, there may be some slight variations due to Nature and/or the tolerances the camera is built to. I shot with a D810.
Here’s a nice yucca shot in 14-bit and properly exposed.
And here’s the same yucca in 12-bit.
I can’t tell a difference.
Next I tried a shot that would test the camera’s dynamic range with sunlit clouds and a shadowed foreground. I exposed not to blow out the highlights, which then resulted in the shadows being pretty underexposed, requiring me to pull the shadow detail back up in post. Here’s the original 14-bit file with no post-processing. (I won’t bore you with the 12-bit original – it looks identical.)
And here’s the 14-bit with the shadows recovered.
And the 12-bit.
Here’s 14-bit cropped in on some shadow detail.
And the 12-bit.
Oh crop, there’s no difference I can tell. I showed this to a photographer with more critical eyes than mine and she couldn’t tell a difference either. Maybe the difference is only visible in a final print. So I printed the two versions and still couldn’t tell a difference.
So far my tests were with well-exposed shots. It holds to reason that if I were to be able to tell a difference it would be in the dark values as when you’re courting the left side of the histogram, you’re dealing with a lot less raw information. After all, in dark conditions such as underexposure, fewer photons are being counted at each pixel site than when it is bright. (See Spencer Cox’s article about the theoretical advantages of having more data to work with by exposing to the right side of the histogram.)
I was driving through Northern Arizona late one afternoon and there were some clouds so I had to detour to The Mittens. Sadly the clouds started shrinking to where I would have a mediocre sunset shot. Rather than pack up and leave, I thought “ah ha”, time to run some more 12-bit versus 14-bit tests. But this time I’ll bracket the exposures and see what happens.
Rather than go through 30 samples of original shots, tweaked in post shots, and tweaked and cropped to 100% shots, lets fast-forward to the most underexposed sample. First the unprocessed 14-bit.
And 12-bit unprocessed.
14-bit tweaked and cropped to 100%.
12-bit tweaked and cropped to 100%.
What’s a guy got to do to see a difference? These are virtually identical, especially given that this was natural light, which isn’t always even and there could be slight exposure variations due to tiny shutter speed and aperture differences. If I can tell any difference at all it might be a teensy bit more contrast in the tweaked 12-bit samples.
The above tweaking was from letting Lightroom auto-tone the images. Not really the best presentation of this file so I went in and tweaked more to my liking and applied the exact same parameters to each file.
14-bit tweaked.
And 12-bit tweaked.
Squinting real hard I’m learning two things. First, I can’t tell the difference between these 12- and 14-bit images unless I look at the metadata. Second, my 24-120 Nikkor is pretty soft in the corners – I should have shot my 50mm.
You may have noticed all of these so far are at base ISO. Maybe I need to go to higher ISOs where the camera will have to amplify the signal hence we might encounter some noticeable differences. I went inside to get low enough light.
Again, instead of running you through dozens of tedious samples, we’ll cut to the chase. These are at ISO 3200. The scene has a ridiculous dynamic range from sunlit bushes outside the window to deeply shadowed pillows inside. First the untweaked 14-bit.
Now the untweaked 12-bit.
The 12-bit looks a tad brighter. The following had identical tweaks save the exposure of the 14-bit was boosted a tad to match the 12-bit. Tweaked and cropped 14-bit.
Tweaked and cropped 12-bit.
Just maybe an eensy bit more constrast in the 12-bit, but I reckon if you printed these out and swapped them around and showed them to me I couldn’t pick out one from the other.
I guess we’ll have get ridiculous and head to that special place where histograms go to die.
Here’s the 14-bit.
And the original histogram. Oh my, even Michael Moore doesn’t expose that far left.
And 12-bits of despair.
Again the 12-bit looks a tiny bit lighter so this is perhaps a 12-bit/14-bit D810 thing, not any variation in light levels or shutter speed/aperture tolerances.
14-bits cropped and tweaked.
12-bits cropped and tweaked with the same parameters.
At last, a difference I can see – the 12-bit file after extreme processing is having some trouble holding highlight detail. However I can see a tiny bit more shadow detail in the 12-bit file. This might be due to the f/16 aperture not closing as far on that particular exposure, hence letting a few more photons reach the sensor. Or maybe not. Either way, both 12-bit and 14-bit files look like they drank too much and puked all over themselves.
Let’s try again. Here’s a scene in Petrified Forest. I’m underexposing heavily and stopped way down trying to get a small, tight sunstar and seeing just how much I can retrieve from the shadows in post. (This would be a much better candidate for exposure blending, but we’re running a test here.)
Here’s the untweaked files. First the 14-bit.
Now the 12-bit.
And now with identical post processing. -100 highlights, +100 shadows, +2.45 exposure, and various other tweaks.
The 14-bit.
And the 12-bit.
At last, I created a practical field test that is showing a difference. The 12-bit version has got an unpleasant greenish cast and the 14-bit is trending slightly magenta. Also it looks like there is more shadow detail in the 12-bit. Let’s fix the color in the 12-bit shot as seen below.
Voila. Looks great now and I like the extra shadow detail recovery. As I’ve already gone to +100 in the shadows there’s no further global adjustments I can give the 14-bit without blowing out my sunstar. I’d have to do local dodging and burning to match the two.
The takeaway I got from all this is that worrying about having 14-bit files instead of 12-bit is silly if you expose well or even just don’t mess up too bad. Good post-processing can give results that make it hard if not impossible to distinguish between 12- and 14-bit files. I did these tests to mimic situations I might encounter. I encourage readers to do their own tests with the sort of subjects they shoot. Feel free to share your results in the comments section.
All said and done, will I switch to shooting 12-bit? Psychologically this tears me apart knowing my files won’t be all they can be. Furthermore I like to photograph birds. They have four cone types in their eyes versus three for humans, hence have far superior color vision, up to ten times better. What if I want to sell family portraits to this Mallard mom?
Us humans can’t see the difference, but Mama Mallard sure will. Aw, those Mallards are pretty stingy anyway. I’ll just switch to 12-bit and increase my chance of getting the shot (through shameless spray and pray tactics) rather than getting a bigger file I can’t appreciate. And I can always switch back to 14-bit when conditions dictate I should – like the next time I’m shooting handheld candlelit test chart shots and forget to remove my 6-stop ND filter.
Text and photos ©John Sherman
Thanks John. Your testing settled the matter for me. All I needed to see. For me it’s 12 bit, lossless compressed raw. Just pre-ordered the new Nikon D500 and will use this setting happy to know that I’ll get a larger buffer and more images on my card and computer. I’ll say this though, after reading some of the other comments: Some people are real hard-headed aren’t they? Maniacal pixel peepers to the grave. Although, perhaps some people can resolve more detail/color/shades than the average person? My vision is 20/20, but my wife is able to read a road sign 2x as far away as I am. She has super normal sight, hearing, and sense of smell. In fact, we kid her by saying that she doesn’t have a nose… It’s a snout! *LOL* So perhaps vision differences can be a factor as well?
Simply if you don’t process the raw photo any bit depth shows the some results.
The typical bit depth are:
The 8bit of jpeg photos.
or the 10bit raw from smartphones like iPhone 7, Nokia 1020,
or the 12bit raw from cameras like the Nikon 1 series or during continuous shooting from Sony a7 series,
or the common 14bit of most dSLR camera and mirrorless cameras,
or the 16bit of nearly all digital medium format cameras.
People don’t forget that raw files are not photo files. Raw files always demand processing before being uploaded to internet or being printed. So always higher bit depth gives greater flexibility with adjustments like highlights, shadows, blacks and whites etc.
The difference sometimes is small. However the difference between photos from professionals or enthusiasts is small too.
You can call my comment “stupid” but that does not help explain or lessen my confusion. Since posting that comment I have actually figured out the source of my confusion and I now understand why more bits in the RAW file are important. And you are correct that is comes down to what you can do in post processing to recover details and how that is achieved.
So yes, I should learn something, and I was hoping that by posting the question I might get a helpful answer. YOUR “helpful” comment added zero to my understanding, and that would be zero in 8-bit, 12-bit, 14-bit… even gazillion-bit format.
You’re a funny guy. Funny, you know, like a clown. A Camera Clown.
MOST or Many monitors are 8 bit.
Windows 7 on up supports 10 bit as do the new iMacs.
Now? Can you SEE the diff between 12 and 14 bit color on a 10 bit montor,
even supposing a proper calibration? I don’t think so, but don’t really know.
I suspect but can’t prove that 14 bit color depth is a sort of ‘future proof’ step
we can easily take today.
As kind of an aside? Go to youtube and listen to high-end loudspeakers. They
ALL sound pretty much the same coming out the nasty little speakers which are
the stock-in-trade of the modern computer. Even when I plug in my good Grados,
I can’t tell. But thru my stereo? You can easily tell the difference between compressed
MP3 (like a compressed jpg) and a Lossless file of the same stuff from the same CD source.
Funny, and as far as I can tell these things, true. Very much like the world of high-bit audio, I suspect, where we are buying massive files only our future computer overlords will appreciate.
Erm… The bit depth encoding has more to do with how many discrete tone values each stop your camera can capture. Higher bit depth = more discrete tone steps for each stop of light captured which means more ability to white balance and generally push the colors around in post before banding and other artifacts show up. Contrary to what people think, the camera’s dynamic range is not linearly encoded in the raw file, so there’s no real correlation between how much dynamic range the file can hold and how many bits it is. For example, Canon 14 bit raw .CR2 files generally encode the sensor data from tone values 2047 to 13586 (roughly, it can vary a bit from one camera to the next, but for the Canon cameras I’ve used, that’s what it is). This has been the case since .CR2 files have existed and doesn’t matter if it’s an old 600D (t3i) which gets just over 11 stops of DR or a new 80D, which gets 13+ stops of DR. The total dynamic range captured is encoded into roughly 11,500 discrete tone steps. Part of the raw development is taking that non-linear raw encoding and linearizing it back out.
You’re experiment has largely bore that out. Given that you’re looking at 8 bits per color on your monitor, given a standard rec.709 gamma (with sRGB), you’ve got 12 stops worth of data encoding into that 8 bit file. As long as you’ve got at least 3 or 4 discrete tones values per displayable tone value, you will not see much difference between the two.
I discovered that there’s one place you will immediately see a huge difference in a 12 bit vs 14 bit photo. Take three or more bracketed RAW files from a D810 through Photomatix for an HDR file. The 12 bit will come out a bright green, no matter what colors were in the viewfinder. 14 bit files work as you would expect. It was driving me crazy until I looked through the FAQs on Photomatix. Sure enough, their current software as of mid Dec, 2016 can only deal with 14 bit RAW files from the D810. JPEGs, should you wish, work just fine, although if you care enough to buy a D810, shoot HDR and go through Photomatix, I would expect you to use RAW.
A very interesting article in which the proof is in the image(s)… I think. And interesting that comments are still coming in a year later.
One point that was made several times is that 24-bit colour depth is all that we can see on the monitor; another point made is that 14-bit data has more information than 12-bit data, hence more room for fine adjustment.
If I put these ideas together, notwithstanding that more bits gives you finer control/resolution over the changes you make, are you not viewing those changes on an 8-bit-per-channel monitor as you tweak your image? Considering just a single colour channel, a change in value from say 1 to 2 in 8-bit space should be visible on the monitor as a change in brightness. If my data structure in computer memory were 16-bit-per-channel it would still get quantized into 8-bit-per-channel for display. Thus, data values from 0 to 255 in 16-bit space would map to 0 in 8-bit space; values from 256 to 511 would map to 1; 512 to 767 would map to 3; etc.
So in doing fine adjustments to my 16-bit data, how would I see the effect of changes in which the 16-bit space value were adjusted from say 585 to 728? Would this change not be invisible, because both values still display as just 3 in 8-bit space? Do I not have to cross the threshold of the mapped 8-bit values before I can detect that I have made a change to my image?
I know this is slightly off the topic of 12- vs 14-bit data, but it has me baffled as to why we need more information than what can be displayed (or printed).
I frequent a number of photography forums and, anytime the issue of image quality comes up, I see pages and pages and pages of text. There are technical explanations that might or might not be correct, there are glossy descriptions of differences noted, but most of all there are pages and pages of text. I enjoy articles like this because they’re image heavy, letting the reader decide whether the reader can see a difference in images. Instead of text, I’d rather see someone in the comments post a dropbox link with visual confirmation of the differences they’re describing … after all, image quality is about images.
The test i a bit misleading. Posted images will not help. Some comments pointed at it already.
14 bit is about more infirmation stored. You will no t be able to see it right away.
If your shot a picture which is perfect for you ( some sharpening and minor color grading might be added)
14 bit is not needed.
But if you start processing the pictures to your liking or to reach a specific look or to rescue a not so well made shot this more of information might help to prevent colorbending or other ugly outcomes.