The Sleeping Giant's Sea Lion

by Jakub

submit your photo


Hall of Fame
View past winners from this year

Please participate in Meta
and help us grow.

Take the 2-minute tour ×
Photography Stack Exchange is a question and answer site for professional, enthusiast and amateur photographers. It's 100% free, no registration required.

A regular JPEG image has only 8 bits to store information about the tone of each pixel. When storing the image in RAW format (for example, DNG), we can store tone using more bits per pixel, which gives us a wider range and more options for processing on the computer.

My current camera can record pictures as 12-bit DNG, and I normally use RAW. I've noticed that newer models of DSLRs are able to store 14 bits per pixel. To me it looks like a huge advantage to get those 2 more bits, but in reality is it a big difference? Would I see the difference in post-processing? Would the difference be more on the darks (underexposed) or highlights (overexposed) parts of the image?

share|improve this question
8  
Small but important correction: JPEGs store 8 bits per channel per pixel: 8 bits of red, 8 of green, 8 of blue, 24 bits in total. Likewise RAW stores 12/14 bits per channel per pixel: 36/42 bits per pixel in total. 8 bits per pixel is what you get with the GIF format (or an 8-bit PNG): a maximum palette of 256 colours which is no good for photos. –  Mark Whitaker Dec 3 '11 at 0:24
3  
+1 to the correction, but note that the raw file really is only storing that number of bits per pixel, where each pixel — each photosite — is filtered to a certain color. Conversion from RAW entails demosaicing, where full-color information is cleverly extrapolated from the neighbors. So RAW isn't really storing 36/42 bits per pixel. (Assuming a standard Bayer or similar color-filter array as is normal for all but Sigma's Foveon cameras.) –  mattdm Dec 3 '11 at 19:46
    
@MarkWhitaker - further correction (even smaller ;-) : RAW (in most cameras) stores 12/14 bit per pixel, of one of the 3 primary colors. The other two colors are added in post-processing by Bayer (or other pattern) demosaicing. An exception to that is sensors of the Foveon type where 3 colors per pixel are captured by the sensor. –  ysap Jun 9 '12 at 15:29
    
Most JPEG images store the image in YCC(en.wikipedia.org/wiki/YCbCr) format. The number of bits for each channel (luminance, delta-blue, delta-red) is determined when the image is compressed. Rendering the image is usually done to RGB888, which leads to a little additional loss(more than the compression) of information. There are also CMYK and YCCK JPEGs, although they are more rare, and there are '12-bit' JPEGs, which render to RGB[12][12][12] instead of RGB[8][8][8] –  rsaxvc Jun 12 '12 at 3:07

3 Answers 3

up vote 13 down vote accepted

It makes some measurable difference but does not tell the whole story. DxOMark's portrait score is a technical assessment of the output of various cameras specifically in terms of color depth, which they carefully describe as having a "correlation" with color sensitivity, which is the actual nuance in color.

If you look at the results of that metric, you can see that the top-scoring cameras have 16 bits per pixel, followed by those with 14 bits per pixel. The expensive medium-format digital backs get DxOMark scores of 24-26 or so, followed by the very top SLRs with a range of 23-25. Then, the cameras with 12-bit/pixels come in next — I think the top one is 22-point-something.

But note that DxOMark describes a difference of 1 in this score as "barely noticeable". That's if you're barely noticing very carefully. For most people, much larger differences in score aren't noticeable either in real-world results.

Impact on the real world and final perception are one reason it's not a huge deal. But there's more! If you go further down the list, you'll find older cameras with 14-bit depth and lower scores than newer 12-bit cameras. So that number alone doesn't tell the whole technical story either. Newer sensor and processing tech improves real results in other ways. If you're comparing current generations, more depth is better, but don't assume that it's everything.

As for whether this gives you more room in the shadows or in the highlights: it's not really that the bits are added at either end — instead, there's just more gradiation. Imagine one newspaper gives movies one to four stars, while another uses a 1-10 scale. A "10" from the second newspaper isn't necessarily a lot better than a four star review from the first, but the additional "bits" allow for more nuance. This is the same idea.

These sensors still suffer from harsh cut-off of highlights, so as always with digital it's best to expose so those are retained and pull detail from the shadow: and yeah, better depth will help that to some degree, if you want to post-process to brighten dark areas, since there will (in theory) be more nuance to stretch out.

An important thing to realize is that the 12 or 14 bits from the sensor, while JPEGs use a gamma curve which fits with human perception. That's not just a way for JPEG to compress data — a curve has to be applied in order for the image to look right. Since this curve does "squish" the bits, that's part of the reason there's less of a perceptual difference than one might expect. (But having that linear data in the un-curved form is part of what gives RAW it's flexibility: it's easy to choose a different curve.)

My overall point, though, is that I wouldn't look at the underlying number to make a decision between two cameras. Instead, look at the final results.


Another external reference, presenting the same point of view, from the American Society of Media Photographers "Digital Photography Best Practices and Workflow" web site's section on sensors:

At the time of this writing [n.b. 2009 or earlier], no 35 mm DSLR cameras that have 14-bit capture ability clearly show an image quality advantage over 12-bit capture.

Some medium-format sensor makers claim an advantage with 16-bit capture. However, we have never seen a study (other than the manufacturer’s) that shows higher bit depth translates into higher image quality based on 16-bit capture alone. In general, the difference between 14-bit and 16-bit capture would not be visible (to humans anyway) unless a severely steep tone curve was applied to the image (on the order of 6-7 stops).

(Emphasis added. Thanks to an earlier answer from Aaron Hockley for the pointer.)

share|improve this answer
    
Great answer! I think you might want to factor in the way bits are divvied up amongst highlights, midtones, and shadows, though. You mentioned that "it's not that bits are added at either end, instead there's just more room for gradation". To my understanding, levels are allocated such that highlights get more, followed by midtones, followed by shadows then darks. Upping the bit depth from 12 (4096 levels) to 14 (16384 levels) SHOULD have an impact on highlights...and a significant one on highlight roll-off when they peak. You have an additional 12,288 discreet luminance levels beyond 12-bit. –  jrista Dec 2 '11 at 20:09
    
Assuming a binary distribution of levels in the upper highlights, lights, midtones, shades, and shadows at 12-bit: 2048, 1024, 512, 256, 256. For 14-bit: 8192, 4096, 2048, 1024, 1024. (A bit contrived, but it demonstrates the point.) That difference should be visible in highlights, particularly with RAW (most raw tools DO apply a tone curve when they import), and result in a smoother, more recoverable falloff before highlights blow out. –  jrista Dec 2 '11 at 20:14
1  
@jrista — turn that thought upside down. The blown highlight problem is an unavoidable hard-stop, having to do with the full-well capacity of the photosites. Higher bit depth doesn't mean bigger wells; it's that the wells are sampled more finely. Since the brighter data already has more bits, adding more sampling there doesn't buy as much as adding more to the more-sparsely-represented dark areas does. –  mattdm Dec 2 '11 at 20:22
    
In your example, going from 2048 to 8192 is 4×, and so is 256 to 1024. But the top level already had thousands of levels of nuance. Consider even further down, where one might be going from 16 possible levels of dark colors to 64 — that's a much more meaningful change than from 4096 to 16384, even though the later is clearly more. –  mattdm Dec 2 '11 at 20:27
1  
I know the blown highlights are a hard problem, however I think when you have more levels to attenuate, it is LESS of a problem. I fully agree with you that having more bits in shadows limits the need to worry about photon counting. –  jrista Dec 2 '11 at 21:12

More bits usually doesn't mean more range, but more precision. That is to say, the ends of the scales, the blackest blacks and whitest whites, will stay where they are (at 0 and the max value) but the number of values between them will be greater with more bits.

You quickly fall into diminishing returns here as there simply is no need for that much precision, and the camera sensor is often not even able to resolve to that precision anyway.

share|improve this answer
    
Yeah. :) This is a nice, much more succinct way of saying the same thing as my answer. +1 and welcome to Stack Exchange! –  mattdm Dec 6 '11 at 21:21

I think there is some confusion related to the differences between 12 and 14 bit RAW when it comes to its impact on the dynamic range.

My understanding is that the 14-bit RAW does not expand the dynamic range. It expands neither the highlights nor the shadows. It gives you more gradual information between the darkest and brightest details the sensor can capture (it is like like you get 4 times more shades of grey). I am pretty sure that I would not notice any difference between 12 or 14 bit raw images captured by the same sensor.

Just for fun have a look at this Colour IQ Test, I am pretty sure it is of less than 12 bit graduality.

share|improve this answer
1  
This color acuity test is great! It (again) proved my point (to myself) on the importance of a good monitor. I have a great Dell U2410 connected to a HP laptop with an average bright screen (true color or whatever they call this awful technology). My browser is open on the laptop desktop and I first took that test on that monitor. Age 40, I got a score of 12, which is not bad at all. Then I decided to do it again, this time on the Dell monitor. My score with a good monitor is now a prefect 0! –  ysap Dec 2 '11 at 19:11
    
Anyway, having more intermediate levels of color is definitely a part of the definition of a dynamic range. Basically, it is the ratio of the maximum value to the minimum measurable unit. –  ysap Dec 2 '11 at 19:14
    
... one more thing - you view this color test on your monitor, which is most probably 10-bit in the best case, with your browser most probably not able to render more than 8 or 10 bits itself, so yes, the test is of less than 12-bit. –  ysap Dec 2 '11 at 19:15
    
I'm not sure it's more than tangentially relevant to the question, but that test is so interesting that it gets a +1 from me in any case. :) –  mattdm Dec 2 '11 at 20:57

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.