Posted by Luke Ormand on October 23, 2009 at 8:17 PM
I just purchased a Nikon D300s and see the option to take 14 bit RAW files. Aside from being larger and taking up more space, it reduces fps from 7 to 2.5. What are the advantages to using 14 bit? If there is another post on this in the recent past feel free to point it out but I didn't find anything.
- Luke
Posted by Bob Miller on October 23, 2009 at 8:57 PM
Posted by Paul DeCesare on October 23, 2009 at 10:00 PM
I don't know the technology, but I do know that I can hit my 14 bit RAW images from my 5D Mark II much harder with Photoshop edits (such as Burn/Dodge) than I ever could with my 12-bit 5D. Having said that, if you are shooting for the web and not large print, even a hard-hit edit goes virtually unseen when you knock and image down to 640 px for web. If you work to get exposure right and aren't printing a four-foot image, I bet you won't notice the difference. Many printers only handle 8-bit as far as output goes anyway.
To complicate matters, Burn/Dodge in PS CS4/ACR 5.X has new algorithms that preserve tones; I didn't use Burn/Dodge much in prior versions. Maybe THAT'S why I get better results. Anyway, a few side-by-side tests should give you your answer, and if it looks good, go for 7fps.
Posted by Greg Maklae on October 24, 2009 at 6:05 AM
With 12 bit, you about 4100 color tonal variations per pixel, while with 14 bit you have just over 16300 tonal variation per pixel, a lot more info.
Posted by David Bostedo on October 24, 2009 at 12:45 PM
The problem with using 14 bit is that, yes, you can do more heavy handed edits, and yes, you get more tones per pixel - BUT the extra bits may be indistinguishable from randomness. In order for 14-bit to be really worthwhile, the noise must be low enough that it's smaller than the fine gradations you get with 14-bit. Otherwise, those extra two bits are swamped by the noise and become random.
Now you have a D300s, which is one of the lower noise cameras out there. It may be possible that in certain cases 14-bit is better, but you're not likely to be able to see it (this is true of the D3, despite its low noise). You can probably get the same advantages just keeping your images in 16-bit when you raw develop. I'd suggest taking a few images at both settings and looking at them closely. Is the difference worth anything to you?
I highly recommend reading this paper entitled Noise, Dynamic Range, and Bit Depth. It's by Emil Martinec, U. of Chicago physics professor and an avid nature photographer, and is written pretty plainly and easy to understand. It's a little out of date - the D3 was still pretty new when it was written - but it still applies.
Posted by Paul DeCesare on October 30, 2009 at 12:35 PM
Luke, this question was in today's (DDQ Digital Darkroom Questions) email from Tim Grey. Tim's answer won't directly answer your specific question but does shed some more light on the subject (wow, no pun intended!):
The actual increase in file size doesn't make any real difference, since that increase in file size is not the result of additional pixels. Instead, it is the result of more decimal places of accuracy (in a manner of speaking) for each of the individual pixel values. So the real question is, is there an appreciable difference between 12-bit per channel and 14-bit per channel color?
In absolute terms, there is a considerable difference. An 8-bit per channel image contains only 256 tonal values per channel, and thus only 16,777,216 million total possible color values. That sounds like a lot (and it is), but it is a small enough number that significant adjustments can lead to posterization (loss of smooth gradations of tone and color within the image). At the top end of "normal" digital photographs is 16-bit per channel, which allows for 281,474,976,710,656 possible color values 65,536 tonal values per channel). With so many colors possible, posterization is highly unlikely even with very strong adjustments to your images.
So, where do 12-bit and 14-bit stack up? A 12-bit per channel image allows for up to 4,096 possible tonal values per channel, and thus a total of about 68.7 billion possible color values. A 14-bit per channel image offers up to 16,384 possible tonal values, for a total of almost 4.4 trillion possible color values. That's a pretty big difference, but frankly in the real world it doesn't amount to a particularly huge difference (at least not anywhere near as big a difference as the numbers presented here would imply).
When applying typical adjustment to an image, you're not going to do enough damage to cause posterization with either a 12-bit per channel or 14-bit per channel image. Because of the additional number of colors available with a 14-bit per channel image it is possible you'll retain greater fine detail within the image. Gradations may be more smooth, and some areas may actually contain more visible detail than if you worked with a 12-bit per channel image. In most cases you would have a very difficult time telling the difference between the two, although in some cases it can make a difference that is clearly visible when you zoom in closely to certain areas of an image.
Despite the fact that there is a very minor benefit in most cases, I do recommend capturing at the highest bit-depth available for your digital camera. Keep in mind that this bit-depth setting affects the accuracy of the analog to digital (A/D) conversion within the camera. The final image you create when you convert the RAW image to a "real" image file format will be either 8-bit per channel (which I don't recommend, as you would be throwing away all the "extra" information you gathered by capturing in RAW) or 16-bit per channel. Photoshop doesn't offer 12-bit or 14-bit per channel file formats, so the 12-bit or 14-bit data will be placed into a 16-bit per channel "package". You don't really have full 16-bit per channel data (though some cameras do offer this support), your data is just held by an image file capable of storing more data than you actually have.
-10/30/2009 DDQ Email, Tim Grey
Posted by David Bostedo on October 30, 2009 at 6:43 PM
That's a good post Paul (although all the talk about available gradations could be a little mis-leading...)
And I agree with his talk of rarely seeing a difference between 14 and 12 bit images. But I don't agree with the statement about the bit depth affecting the accuracy of the A to D. No matter how many bits you add there, it doesn't add to the accuracy of the A to D conversion unless the noise involved is less than the accuracy of the bits you're adding. In the huge majority of cases, that noise isn't less, and you are just adding random data. There may be a slight benefit in having some random data there (a little noise is actually a good thing, especially for post-process) but it doesn't create real data if noise ends up making it random values.
Discussion
6 replies- Luke
I seemed to have seen a long post on the topic recently on FM, so I did a search and here it is- http://www.fredmiranda.com/forum/topic/810505/
To complicate matters, Burn/Dodge in PS CS4/ACR 5.X has new algorithms that preserve tones; I didn't use Burn/Dodge much in prior versions. Maybe THAT'S why I get better results. Anyway, a few side-by-side tests should give you your answer, and if it looks good, go for 7fps.
Now you have a D300s, which is one of the lower noise cameras out there. It may be possible that in certain cases 14-bit is better, but you're not likely to be able to see it (this is true of the D3, despite its low noise). You can probably get the same advantages just keeping your images in 16-bit when you raw develop. I'd suggest taking a few images at both settings and looking at them closely. Is the difference worth anything to you?
I highly recommend reading this paper entitled Noise, Dynamic Range, and Bit Depth. It's by Emil Martinec, U. of Chicago physics professor and an avid nature photographer, and is written pretty plainly and easy to understand. It's a little out of date - the D3 was still pretty new when it was written - but it still applies.
-10/30/2009 DDQ Email, Tim Grey
And I agree with his talk of rarely seeing a difference between 14 and 12 bit images. But I don't agree with the statement about the bit depth affecting the accuracy of the A to D. No matter how many bits you add there, it doesn't add to the accuracy of the A to D conversion unless the noise involved is less than the accuracy of the bits you're adding. In the huge majority of cases, that noise isn't less, and you are just adding random data. There may be a slight benefit in having some random data there (a little noise is actually a good thing, especially for post-process) but it doesn't create real data if noise ends up making it random values.