Digital Media | Spotlight: Photography | Inside Lightroom | Blogs
14-bit capture and Lightroom
The latest batch of digital SLRs on the market are fitted for the most part with 14-bit sensors. There have been a few websites out there running tests on the difference between 14-bit and 12-bit capture. And while some have been reporting that there isn't much if any difference the reality is that a 14-bit file is a vastly superior to a 12 bit image file. I'll explain why in just a bit but let's explain what 12-bit and 14-bit capture means and what the differences are.
The bit depth of a digital image sensor refers to the number of tones (or in effect colors) that the camera can record. First, remember that all digital sensors record images in black and white - color is then interpolated by the camera and software because each pixel has a color filter over it known as a Bayer pattern.
So, if we just look at the grayscale tones it will be easy to understand the bit depth math. Basically a bit depth of 1 would be 2 to the 1st power= 2 possible colors, either black or white. Hence a bit depth of 12 would be 2 to the 12th power = 4,096 possible tones and a bit depth of 14 would be 2 to the 14th power = 16,384 possible tones. So what does this mean? Basically with 12-bit capture we have 4,096 possible colors per channel (and there are 3 channels: red, green and blue) and with 14-bit we have 16,384 possible colors per channels. That is a big difference. It means that the 14-bit sensor will be able to delineate a much finer gradation of colors than the 12-bit sensor.
Now this doesn't mean that a 14-bit sensor will have larger exposure latitude than a 12-bit sensor. That depends on other factors including the size of the individual pixels, in-camera processing and how you have your camera set up - i.e., contrast settings, etc.
In terms of Lightroom, what a 14-bit sensor gives you that a 12-bit cannot is a little extra room to tweak your images without the image falling apart. If you have ever really started cranking sliders to extremes in the Develop module you will notice - if you look at your image at 100% - that the image starts to get grainy or noisy looking. It starts to fall apart. This is because you are severely altering the data that was captured by the camera. The more radically you adjust the sliders in Lightroom the more the image quality (on the pixel level and beyond) degrades. If you take it to extremes it becomes very noticeable. This is because you only have so many tonal values to work with. Hence, if you have more tonal steps, as with 14-bit images, then you can go to greater extremes with your processing before you'll start to see the image degrade.
Of course if every image you shoot is perfectly exposed then this isn't that big of an issue. But if you want to apply that cool "300 movie" look to your images at some point the quality will decline. By how much depends on how far you adjust the image. The difference is most noticeable in the shadows since less information is recorded for the shadows than in the highlights on every sensor no matter what the bit-depth. And it is very noticeable in the way highlights fall off. In 14-bit you have a finer gradation so there is a much nicer (i.e. more film like) transition in the highlights from pure white to off white and beyond.
In general, I am very excited to be shooting with a 14-bit Nikon D300 (and a D3 to be coming soon hopefully). 14-bit capture was a major reason I bough the D300 and I hope that soon enough we'll have true 16-bit capture. The files are a little bigger and the cameras need ever-larger memory cards to deal with it but when you are working up images the extra tonal information really makes a big difference.
That's it for this session. See you next week.
Adios, Michael Clark
So do you have any side by side examples of 14-bit files being superior to their 12-bit equivalents? 12 bits are enough to capture the full dynamic range of DSLR sensors. I suspect that the ACR engine works with 16-bit data internally although only Adobe engineers can confirm that. You are mixing up the in-camera data and the working space during post processing - where it is accepted that 16-bit ProPhoto RGB is the way to go. Differences between 8-bit and 16-bit (during processing) are easily demonstrated, but until I see a side-by-side example of 12-bit vs 14-bit raw files I remain sceptical.
Unless I'm missing something (possible), the 14 bits don't help unless there is actually 14 bits worth of data coming from upstream. The difference is big: 4,096 levels vs. 16,384. Have sensor sensitivities quadrupled?
Explaining the differences in the number of tonal steps is just step #0 in understanding 12- vs. 14-bit. It's the obvious step. Once you understand this, however, you need to understand that the signal entering the analog-to-digital converter is a noisy signal. The process of going from analog-to-digital can be viewed as adding an additional component of noise known as quantization noise, which is basically the difference between the original signal and the discrete step assigned by the A/D converter. This quantization noise will obviously be less for a 14-bit converter compared with 12-bit. However, if you add two sources of independent noise and one is much larger than the other, the larger component will dominate. In almost every single case, the noise in the analog signal entering the A/D converter is much larger than the quantization noise added by a 14-bit or even a 12-bit (except in very few DSLRs, and only at the base ISO) converter. As a result, the advantage of 14-bit over 12-bit becomes virtually irrelevant. This has been bourne out by some users taking a 14-bit raw data file and replacing the least significant 2 bits with the fixed binary pattern "10" (a midway value) to turn it into a 12-bit file. I've yet to see a study that didn't find it impossible to see difference between images processed from the original and this created 12-bit version.
Now, there is a secondary advantage of moving to 14-bit conversion. Higher bit-depth converters tends to be higher quality in general because they need to be linear to 1/2 a bit in order to get their 14-bit rating. Thus, they can result in better data even when their least significant bits are ignored. Further, their use sets the processing pipeline and file formats in place for future sensor improvements that might be able to leverage the extra bits. For now, however, we're really just getting 11- to 12-bit quality data out of our 14-bit converters ... and data which is not only larger but also less losslessly compressible due to the random nature of the extra bits.
Possibility of samples to back up the claims?
Regards,
Richard :)
"Possibility of samples to back up the claims?"
I can point to a variety of threads on other forums but I think I'll create some test images of my own to post. As an engineer with a technical background in analog and digital signal processing, it's clear to me in principal and from the way we design any such system that the required bit depth of the ADC at the end of an analog signal chain depends on the signal-to-noise ratio of the signal. Once the noise is much larger in magnitude than the size of the spacing that moves you from one digital value to the next, the value of the extra bit depth is lost.
However (and this may be a big however), there is still benefit to a high bit-depth ADC beyond the argument given in original article. In fact, this benefit occurs *because* of the argument that I make above rather than despite it. The benefit is that when you have a high-bit depth ADC, you don't have to amplify the incoming analog signal as much. Normally, you'd amplify this signal to the point where the range of analog values just about fits the input range of the ADC so that you can take full advantage of all the available tone values. If you have way more than enough tone values, you can get away with not filling the ADC's full input range. For example, if the maximum value of your analog input signal only reaches half the maximum input value of the ADC, you effectively lose a bit of depth (half the number of tone values). If you have 14 bits, however, and need only 12, you can get away with this. The advantage is then that you don't need a different analog gain for each ISO that you wish to implement, saving on implementation complexity and cost. The K10D, with its 22 bit ADC, takes advantage of this. It has only *one* analog gain and thus doesn't have different ISO gains like in other cameras. All ISOs are implemented by digital scaling after the ADC step. Since there is so much excess bit depth, however, this is no problem. Creating ISO 3200 from the same analog gain as ISO 100 means losing 5 bits of bit depth, but that still leaves you with 17 bits of information from the ADC ... enough to create a 12- or 14-bit raw output file. So, there's an engineering advantage here if not an advantage to the end user.
David
Sorry I have been out of the loop. I am still in Chile and will be back in the office on Tuesday then i can get around to replying to these comments while I process the thousands of images I have shot this past three weeks. Hope all are well and thank you for the feedback...
Cheers, Michael
Ok, I am back in the office but buried under a pile of work - I'll try to at least address some of these comments here this week. Thanks for all of the comments and the feedback!