Home · Register · Search · View Winners · Software · Hosting · Software · Join Upload & Sell

Moderated by: Fred Miranda
Username   Password

  New fredmiranda.com Mobile Site
  New Feature: SMS Notification alert
  New Feature: Buy & Sell Watchlist
  

FM Forums | Nikon Forum | Join Upload & Sell

1       2      
3
       4       5       end
  

Archive 2012 · D800: 12-bit RAW vs. 14-bit RAW
  
 
taob
Offline
• • •
Upload & Sell: Off
p.3 #1 · p.3 #1 · D800: 12-bit RAW vs. 14-bit RAW


mshi wrote:
14-bit offers 16,384 different possible steps while 16-bit does 65,536 possible steps at sensor capture level. In other words, 16-bit sensor has four times the fidelity (or precision) of the 14-bit sensor. 65,536 tonal values offer much smoother gradient transition and shadows.


In theory, yes. But here in the real world, the practical difference is often zero. Your monitor cannot show 65536 discrete levels (maybe 1024 at best). You cannot print 65536 discrete levels. In fact, your eyes and brain cannot distinguish anywhere close to 65536 discrete levels at the same time.

In addition, we live in an analog world. Just because you have 16-bit data in a file does not mean that your source is a clean 16-bit signal. Image capture is still an analog process. Photons land (or don't land) on a particular photosite based on statistical probability. The quantity of photons is converted into an electrical signal, which in itself is not a 100% accurate process either. So while 16 bits is a lot of precision, unfortunately the devices we use to capture that data are not that precise. This imprecision shows up as noise in our images. Or at least, it shows up in the image data... in many cases our eyes are not good enough to see the noise.

You can read more about this here, and the first paragraph of that page basically mirrors what I just said above:

http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html

But that reminds me what Leonardo Da Vinci once famously said, " There are three classes of people: Those who see. Those who see when they are shown. Those who do not see."

You forgot the fourth class of people: those who think they see what they are seeing, but aren't really seeing what they think they are seeing.



Apr 11, 2012 at 02:31 AM
taob
Offline
• • •
Upload & Sell: Off
p.3 #2 · p.3 #2 · D800: 12-bit RAW vs. 14-bit RAW


wfektar wrote:
Perhaps the D800 has a low enough noise level and a high DR that if you have a contrasty situation using the entire DR and need to lift the shadows, 14 bit will have a perceptible advantage over 12 bit -- but that remains to be demonstrated. (Any D800 owners care to try?).


I do have some sample images that visually show at what point 12-bit vs 14-bit makes a difference, but I'm hoping for some more interesting subject matter first. I may post my example anyway, which is a bunch of M&Ms in the studio.



Apr 11, 2012 at 02:33 AM
wfektar
Offline

Upload & Sell: Off
p.3 #3 · p.3 #3 · D800: 12-bit RAW vs. 14-bit RAW


taob wrote:
I do have some sample images that visually show at what point 12-bit vs 14-bit makes a difference, but I'm hoping for some more interesting subject matter first. I may post my example anyway, which is a bunch of M&Ms in the studio.


That would definitely be interesting!



Apr 11, 2012 at 02:44 AM
taob
Offline
• • •
Upload & Sell: Off
p.3 #4 · p.3 #4 · D800: 12-bit RAW vs. 14-bit RAW


Here's a visual example of how bit depth often matters much less than people might think. I've taken the 100% crop of mshi's image and successively discarded half of each channel's values, then dithered the image. Here's the original:





Reduced to 7 bits:





Reduced to 6 bits:





Reduced to 5 bits:





Reduced to 4 bits:





Reduced to 3 bits:






To me, it does not become objectionable until around 5 bits (32 levels). That's after throwing out 87.5% of the data from the 8-bit version. Compared to the original 16-bit data, over 99.95% of the data has been discarded. Keep in mind that this is a tiny crop from a 40-megapixel image. The differences will be even less apparent when viewing the photo in its entirety.

The 6-bit image is pretty good, and the 7-bit image looks just as good as the original. I have to zoom in to 300%, flip between the 7-bit and 8-bit versions, and stare at individual pixels just to perceive a difference. If someone just showed me the 7-bit version without saying anything, I would not suspect it was anything but a full-fidelity image.



Apr 11, 2012 at 02:59 AM
Taoguy
Offline
• • •
Upload & Sell: On
p.3 #5 · p.3 #5 · D800: 12-bit RAW vs. 14-bit RAW


taob
You forgot the fourth class of people: those who think they see what they are seeing, but aren't really seeing what they think they are seeing.


Thanks for the excellent posts and explanation.



Apr 11, 2012 at 03:08 AM
taob
Offline
• • •
Upload & Sell: Off
p.3 #6 · p.3 #6 · D800: 12-bit RAW vs. 14-bit RAW


afm901 wrote:
If there were not a noticeable difference between 12-bit and 14-bit, Nikon would not offer 14-bit as an option.


There is no noticeable difference in the large majority of cases. However, in certain circumstances, there can be a difference. Whether one ever needs to post-process your images under those circumstances is dependent on the photographer. In order to judge your need for 12-bit vs 14-bit (or lossy vs lossless), you need to know the facts, not go by other people's opinions. After all, they may have different requirements than you.

To your second point, why does Nikon still offer an uncompressed NEF option on the D800, D4, etc.? By your logic, there must be some benefit to it, otherwise Nikon would have eliminated the option, right?

I can't understand the logic of using 12-bit if file size is your issue with it. Storage is relatively inexpensive.

It isn't only a matter of absolute storage cost, but storage efficiency. If there really was no consequence to file size, we would all be saving everything as 16-bit PSD files, and there would not be a single JPEG to be seen anywhere. And yet JPEGs are extremely popular. I bet even you use them. Why? Because it is a highly efficient method of storing digital photos: you save a lot of disk space for very little reduction in image quality. If I can reduce my storage requirements by half, that would make sense regardless of how little storage costs.

Not all storage is cheap either. I edit my work from a 256GB SSD, which cost me about $300 last Thanksgiving. If my storage requirements suddenly tripled (e.g., going from the D700 to a D800), I may need to replace that SSD with a 600GB one. That's a $1000 expense. If there is a way I can continue to work with my current SSD without noticeably compromising image quality, it would be worth my time to investigate.

Don't forget about other factors too, not just storage capacity. Transmission times will increase as well. A 72 MB uncompressed NEF will take over twice as long to download as a 30 MB compressed NEF. If you are using an online backup service, your file transfer times just doubled or tripled. In most cases, it costs a lot of money to increase your network bandwidth, assuming it is possible at all. A backup that used to comfortably run overnight while I sleep now starts hogging my connection during the day, or else I have to split up the transfer over two or more nights.


Edited on Apr 11, 2012 at 03:17 AM · View previous versions



Apr 11, 2012 at 03:15 AM
taob
Offline
• • •
Upload & Sell: Off
p.3 #7 · p.3 #7 · D800: 12-bit RAW vs. 14-bit RAW


Taoguy wrote:
Thanks for the excellent posts and explanation.


Great name, btw.



Apr 11, 2012 at 03:16 AM
taob
Offline
• • •
Upload & Sell: Off
p.3 #8 · p.3 #8 · D800: 12-bit RAW vs. 14-bit RAW


JimFox wrote:
That is cool to see the difference. What I didn't know was how lossless vs lossy would affect the highlights. And even if it's only slight, there are times where the increase in file size would be worth it. Thanks for these examples!


My pleasure! As I mentioned earlier, there is no question that differences exist between 12-bit vs 14-bit, and lossless vs lossy. The decision each person should make is whether those differences are relevant to the way they shoot and to the type of photography they are creating. And to make a good decision, you need to have the facts. If you don't know how 12-bit and 14-bit images are different, then you cannot make an intelligent decision about them. I'm not trying to tell anyone which method they should choose. I just want to post examples that will hopefully help guide people to the right decision.



Apr 11, 2012 at 03:22 AM
ausemmao
Offline
• •
Upload & Sell: Off
p.3 #9 · p.3 #9 · D800: 12-bit RAW vs. 14-bit RAW


taob wrote:
Can you see any differences? Are you able to tell which images are shot at what settings? This is the point many people have made in the past: there is a mathematical difference between the various compression options, but you are not going to notice 99.9% of the time. I figure I'll use this as a starting point, then move on to situations that do show the differences.


In fairness, ISO6400 isn't going to show the difference between them because 14 or 12 bit, it's mostly recording noise (which reinforces your point).

mshi wrote:
14-bit offers 16,384 different possible steps while 16-bit does 65,536 possible steps at sensor capture level. In other words, 16-bit sensor has four times the fidelity (or precision) of the 14-bit sensor. 65,536 tonal values offer much smoother gradient transition and shadows. Whether or not I can tell the difference all the time, I don't know. But that reminds me what Leonardo Da Vinci once famously said, " There are three classes of people: Those who see. Those who see when they are shown. Those who do not see."


The problem with this is that there simply isn't the gradation in the incoming light to make the bit depth useful in highlights. Noise in light rises with the square root of the quantity captured. Consider a high FWC of 100k e-, with 100% QE for simplicity's sake. Your highest stop has 32,768 possible values, but between 50,000 and 100,000 there are less than 200 discernable values. With a signal of 50-100k, the shot noise is 200-330 - a recorded value of 70000 could have 'really' been anywhere between 69850 and 70150.

So those extra values provide dithering from noise, and far more than needed to prevent posterisation. Even the 200 discrete distinguishable values are more than enough given the nature of eyesight and display technologies . Accounting for a healthy dose of dithering, you could throw away over 28000 of those 32768 bits and still have all the 'real' data. For 14 bit, it's over 12000 bits you could throw away and still have near no visible difference in the ability to recover highlights.

Obviously this is different with shadow detail - there the increased bit depth does come in useful, if significant tonal manipulation is carried out, as with a linear encoding system you'd be able to record more useful information. But lossy compression doesn't affect shadow detail anyway, so no loss there.

I'd be willing to bet a lossy compressed 16 bit capture (so lossy that it was smaller than a lossless compressed 14 bit capture) would be more amenable to being tortured in post than said 14 bit file, but the problem with that is that you need very fast 16 bit ADCs - and given that even MFDBs barely justify 16 bit capture now, it's probably not happening any time soon.



Apr 11, 2012 at 12:12 PM
penpro
Offline
• • •
Upload & Sell: Off
p.3 #10 · p.3 #10 · D800: 12-bit RAW vs. 14-bit RAW


Past weekend I shot half the time 14 bit on my D7000 and then switched to 12 bit as I wanted to see if I could get more images in the buffer as I was trying to get birds taking off. I think that I was getting an extra shot or two in a sequence. Once processing in PS I saw no real difference when trying to save an image because of exposure issues.

I will probably leave the camera on 14 bit until I need that extra frame as drive space is cheap and I have about 10 terabytes on the system now.

I think that the other way to save drive space besides not taking as many shots is deleting the ones that obviously will never be used. Every time I go through a days shooting I start by deleting what is out of focus or just an all round bad shot. Unfortunitly for me that can be quite a number.



Apr 11, 2012 at 01:20 PM
 

Search in Used Dept. 



taob
Offline
• • •
Upload & Sell: Off
p.3 #11 · p.3 #11 · D800: 12-bit RAW vs. 14-bit RAW


ausemmao wrote:
In fairness, ISO6400 isn't going to show the difference between them because 14 or 12 bit, it's mostly recording noise (which reinforces your point).


Yeah, the 3x3 grid of images was something I had prepared for another forum. Somebody implied that 14-bit lossless NEFs are needed particularly for high ISO images. The thinking was that you don't get good DR or smooth gradients at high ISO settings, so shooting at the highest possible quality will somehow offset that effect. This sort of misunderstanding seems to be fairly common, which can cause people to make the wrong decisions.

I have an ISO 6400 version of the sky banding shots (which shows no apparent banding even with 12-bit lossy), but I'm afraid someone out there will misinterpret the data and claim that I'm telling everyone to shoot bright outdoor scenes at ISO 6400.




Apr 11, 2012 at 02:41 PM
mshi
Offline
• • • •
Upload & Sell: Off
p.3 #12 · p.3 #12 · D800: 12-bit RAW vs. 14-bit RAW


taob, thanks for the enlightenment by showing the same image at different bit depth. Actually, I was wondering why Nikon hasn't given us just a plain and simple bitmap mode for those of us who hate to do round trip in PS to get the same effects.


Apr 11, 2012 at 06:09 PM
taob
Offline
• • •
Upload & Sell: Off
p.3 #13 · p.3 #13 · D800: 12-bit RAW vs. 14-bit RAW


mshi wrote:
taob, thanks for the enlightenment by showing the same image at different bit depth. Actually, I was wondering why Nikon hasn't given us just a plain and simple bitmap mode for those of us who hate to do round trip in PS to get the same effects.


It is probably something they could add to the Retouch menu, but the end result would be a JPEG, not a raw file. Nikon doesn't seem to like mucking around with their raw files, the way Canon does with the different-sized raw-but-not-really-raw CR2 files.



Apr 11, 2012 at 07:30 PM
mshi
Offline
• • • •
Upload & Sell: Off
p.3 #14 · p.3 #14 · D800: 12-bit RAW vs. 14-bit RAW


taob wrote:
It is probably something they could add to the Retouch menu, but the end result would be a JPEG, not a raw file. Nikon doesn't seem to like mucking around with their raw files, the way Canon does with the different-sized raw-but-not-really-raw CR2 files.


If one is only interested in 1-bit per pixel (bitmap), why bother with RAW in the first place?



Apr 11, 2012 at 07:35 PM
taob
Offline
• • •
Upload & Sell: Off
p.3 #15 · p.3 #15 · D800: 12-bit RAW vs. 14-bit RAW


Here are a couple of examples from the other end of the scale: the effect of 12-bit vs 14-bit on rendering of shadow details and colour. Recall that lossless vs lossy affects highlights. 12-bit vs 14-bit affects shadows. Download the 100% sample crops here:

http://luxography.ca/Images/FM/12vs14.zip

I setup a test shot with random objects from my kitchen. There is a lot of fine detail to capture, so we'll see what happens if those details need to be pulled out from deep shadow. ISO was set fairly low at 400, and I adjusted the aperture and flash power to vary the exposure. Here's how the objects look under good lighting:





Here is an unadjusted image under test conditions, six stops underexposed:





Using a combination of the Exposure slider and Curves tool in ACR 7.0, I brought both the 12-bit and the 14-bit images back up to the proper exposure levels. The full-sized images are in the ZIP file above:

12-bit:





14-bit:





The first thing I'll say is that for being 6 stops under, both images are surprisingly usable. There was no NR applied, so you might want to do some of that first, but otherwise if you had nothing else to use, these could work in a pinch. Pushing ISO 400 by six stops puts you at around ISO 25600. However, we are now in the realm of those 2 bits of difference between 12-bit and 14-bit. Because there aren't as many levels available, colour accuracy in the deep, deep shadows may be compromised. You can see this in the tone of the OOF background areas (much grainier green channel), as well as dark-coloured areas like the marmalade jar, and the lid of the yellow jar of baking powder. Compared to the properly-exposed shot above, you can see that the colours in the 14-bit version are more accurate.

But as you move away from the deep shadows, colour reproduction becomes less of an issue. In both cases, the noise from the signal amplification is the real problem, and that doesn't change much with bit depth in the higher tonal ranges. Looking at the 100% images in the ZIP file, fine detail seems to be equally good (or equally bad ) in both versions. The microprint pattern on the $20 bill, the fine print on the baking powder and marmalade jars, etc. are equally readable.

So there is a difference in deep shadow colour accuracy between 12-bit and 14-bit, but not as much difference in terms of image detail. Back to the question: is it worth shooting in 14-bit for this reason? If you often find yourself needing accurate colour in areas underexposed by 4, 5 or 6 stops or more, then perhaps you are better off with 14-bit capture. For me, I rarely need to rescue a shot by more than 2 stops. I would say that if you were truly concerned about colour reproduction accuracy, taking the time to get the correct exposure or bracketing your exposures will yield much, much better results. The quality from a good 12-bit file far exceeds that of a poor 14-bit file that needs rescuing in post.



Apr 11, 2012 at 07:53 PM
taob
Offline
• • •
Upload & Sell: Off
p.3 #16 · p.3 #16 · D800: 12-bit RAW vs. 14-bit RAW


Here's another example (contained in the same ZIP file in the previous post) using a bag of M&M's I keep in the studio to bribe kids. Same deal, underexpose by 6 stops, then bring back in ACR:







ACR 7.0 only lets me push 5 stops, which is not quite enough to get back to the right exposure:

12-bit:





14-bit:






Using the Curves tool gets me the rest of the way:

12-bit:





14-bit:






Again, the green shift in the darker parts of the image is evident. Zooming in on different parts of the image, you can see that the 14-bit image resolves colours deeper into shadow than the 12-bit image. Whereas the 12-bit image might only have 3 or 4 shades of red to dither before clipping to black, the 14-bit image might have 10 or 12 shades to work with. There are a few spots that appear to be noisy black in the 12-bit image, but actually show some noisy red or blue in the 14-bit image. Highlights are again not affected as much, at least until they roll off into the shadows.

The shadow areas I've zoomed in on are so dark that in the unadjusted exposure, the RGB pixel levels are either 0, 1 or 2 out of 255. So you're looking at areas below 1% luminance, and trying to scrape some sort of usable image detail out of that. I've exaggerated things here for the sake of demonstration, so while you don't necessarily have to go as far as 6 stops of underexposure to see a difference, I don't think anything above, say, 4 stops of underexposure will see much benefit.

Again, if you often need to pull colour detail from areas of an image underexposed by 4 or more stops, then you may have a case for leaving the camera in 14-bit mode. Otherwise, the difference in bit depth simply does not translate to an appreciable difference in the final image. Of course, you are free to use your camera however you see fit, but I wanted to lay to rest some of the myths and misconceptions about 12-bit vs 14-bit and lossy vs lossless compression that I see all the time.



Apr 11, 2012 at 08:09 PM
taob
Offline
• • •
Upload & Sell: Off
p.3 #17 · p.3 #17 · D800: 12-bit RAW vs. 14-bit RAW


mshi wrote:
If one is only interested in 1-bit per pixel (bitmap), why bother with RAW in the first place?


1-bit image? You mean a monochrome bitmap that only uses black or white pixels? Aside from people who want the look of the original Macintosh, who would be interested in that?



Apr 11, 2012 at 08:11 PM
afm901
Online
• •
Upload & Sell: Off
p.3 #18 · p.3 #18 · D800: 12-bit RAW vs. 14-bit RAW


Brian,

Just because you can't easily quantify the visual differences between a properly exposed 12-bit and 14-bit image does not mean the differences are not there. In other words, just because you can't point to an area of an image see an obvious difference does not mean there is not a subtle but real difference that makes the 14-bit image better. This is what all of your testing is missing. It's not a myth or misconceptions either....

If you want to use 12-bit lossy, go for it. I'm not......

Scott



Apr 11, 2012 at 09:28 PM
taob
Offline
• • •
Upload & Sell: Off
p.3 #19 · p.3 #19 · D800: 12-bit RAW vs. 14-bit RAW


afm901 wrote:
Just because you can't easily quantify the visual differences between a properly exposed 12-bit and 14-bit image does not mean the differences are not there.


Actually, yes it does mean exactly that. The limitations of our display technology, our choice of file format (e.g., JPEG), and indeed our frail human vision make it impossible to measure the difference between a properly-exposed image recorded with 12 bits of resolution vs. 14 bits of resolution, at least with the recent Nikon models I've tested. It is only when we need to manipulate the data in fairly extreme ways (e.g., to correct for underexposure, or for artistic purposes such as oversaturation) that the differences manifest themselves.

You should study how Bayer demosaicing works, as well as the mathematical relationship between linear raw data and gamma-corrected image data, and you'll understand how differences in intensity at the 12-bit or 14-bit resolution can map to the same, single 8-bit value displayed on your monitor and seen by your eyes.

If you are still not convinced, then go back to post #18 in this thread and take up the challenge I have there. Download the image at 3x3 test image and tell me which squares were shot in 12-bit lossy mode and which were shot in 14-bit lossless mode. As per your criteria, those are properly exposed images. If the differences are there, then I expect you to have a 100% success rate. I'll even allow the use of mechanical and electronic aids, if you don't think your own eyes are up to the task. Even if you randomly guess, you'll still have a 1-in-512 chance of getting them all correct. Get back to me with your answers, please.



Apr 11, 2012 at 09:46 PM
theSuede
Offline
• • • •
Upload & Sell: Off
p.3 #20 · p.3 #20 · D800: 12-bit RAW vs. 14-bit RAW


Of course there is a REAL difference. The real difference is 2 bits. That's not the question; the question is:
-"Is this (admittedly real) difference discernible, visible or even relevant?"

-Outside the fringe examples where you need lots of shadow lifting or extreme highlight recovery of contrasts in strong color.

Thank you for the samples, Taob, they once more show what most people relying on actual side-by-side inspection of equal circumstance shots have known since Nikon introduced their first version of the "gamma-curve" compression (the newer form - that was introduced with the D300 I think - has a slightly different linear part, that seems to fit the sensor noise statistics better). It really needs to be reiterated once in a while. And it's good to do it with a new camera, to "get to know" its' strengths and weaknesses.

[edit/add - a bit late on the case, but anyway :-) ]



Apr 11, 2012 at 09:47 PM
1       2      
3
       4       5       end




FM Forums | Nikon Forum | Join Upload & Sell

1       2      
3
       4       5       end
    
 

You are not logged in. Login or Register

Username   Password    Retrive password