Bit depth/Colour reproduction, resolution & dynamic range (5 replies and 1 comment)
I'm currently writing my dissertation on camera sensors and I've outlined bit depth/Colour reproduction, resolution & dynamic range as being the most important components of creating a quality image - broadly speaking. It would be much appreicated if you could confirm or deny this.
Also, there seems to be great similarities in bit rate and dynamic range, I understand they are measured differently but do they share the same information?
Keep up the good work,
I think you'll need to talk to an engineering type to get a more accurate description that what I can provide.
There is a rough correlation between bit depth and dynamic range but it doesn't have to be a 1:1 situation. After all, you can display more than 8-stops of information in an 8-bit video signal, which most broadcast video is. It's just that at very low bit depths, you risk having banding artifacts within gradients of tone, like in a clear twilight sky.
In a purely linear raw recording situation, again, there is a loose correlation between number of exposure stops and bit depth. But if you look at something like the Alev sensor in the Alexa camera (and I must preface this with saying that I'm giving a probably inaccurate layman's understanding so don't quote me), you have dual A/D outputs for each photosite, each A/D is a 14-bit processor, but the signal is combined into a wide dynamic range image (14+ stops) into a 16-bit linear file... but then Arriraw records this as a 12-bit Log file (not log though as in the Log-C gamma curve used after debayering to RGB). Arriraw could have been a 16-bit container but the last two stops recorded would be mostly noise at the floor of the signal, so I guess they settled on uncompressed 12-bit log as the most efficient way of storing the 14+ stops of raw data.
Outside of raw recording, any gamma curves applied to the RGB image allow more stops of dynamic range to be recorded in a smaller bit depth range, but at the risk of banding artifacts if you go too far. Hence why people keep pushing the still cameras that can shoot log gamma video to abandon 8-bit for 10-bit recording (something the Panasonic GH5 offers, for example.). It's not that an 8-bit video camera gives you 8-stops of DR and a 10-bit video camera gives you 10-stops of DR. In fact, many digital cinema cameras like the Panavision Genesis recorded about 12-stops of DR into a 10-bit log format.
Great, thanks for your reply. Much appreciated.
Hi JJ, awesome dissertation subject !
I had roughly the same subject a couple of years ago.
I'd be happy to help you if you have any questions regarding DR and sensitivity.
There is three phases the signal goes through that limit the Dynamic Range, the second one being the sensor's Bit Depth. (first one is pixel size)
Pretty much : this "DR2" = bit depth.
Take the number of possible values you can record on.
You know that everything is linear at that stage.
This means that removing a stop of lights divides the value by 2.
And adding one stop doubles the recorded value.
So you can ask :
how many times can I divide my max value by 2 before I reach my lowest value ?
The mathematical translation of this english question is,
what is the log2 of my max value ?
Well.. as it happens
log2(max val) = log2(2^bit depth) = .. you guessed it.. bit depth.
(log2(x) and 2^x cancel each other out)
So the DR will always be inferior to the sensor bit depth.
And the follow up question is.. do you include the last stops recorded in what you consider dynamic range ? Because you're recording whole stops of light with just a couple of values.
I made a couple of videos» on this subject a couple of years ago. I need to stop procrastinating and do the follow up on ISO, log, raw, HDR, gamut size etc...
So the DR will always be inferior to the sensor bit depth
Unless you use an HDR technique like the Alexa's DGA that is.
(2*14= virtually 16 bit)
Keep in mind the difference in the bit depth of the A/D processor on the sensor and the bit depth of the recording format. You can record more than 8-stops of DR in an 8-bit format.
As a said before, Alexa raw is 12-bit log for a 14+ stop image.