DVC PRO color bit depth

[Gopher_Greene]

I'm trying to get an understanding of how color is manipulated from cradle to grave. So begining with capture and the HVX 200, after digging through the manual and other research all I can come up with is quantitizing 8 bit. So if the camera is outputing 8 bit color what does it matter what the color bit depth of the NLE, as long as its 8 bit. The new Premiere CS3 is 10 bit, Avid I think is 10 bit, FCP is 8 bit unless you have uncompressed 4:4:4 then it is 10 bit and Vegas is 8 bit. IS there any benefit in 10 bit color depth in the NLE. Secondly even if you can get 14 bit out of the camera what will it matter if you only have 10bits in the NLE?

[THoff]

10bit is useful once you start doing color correction -- it can encode four times as many discrete values as 8 bits can. Once you start shifting colors around, having ten bits gives you much finer control.

Even if you output to an 8bit format in the end, being able to retain 10bits while you are in the editor is beneficial because the loss doesn't happen until you output -- every color correction, filter and effect you apply in the editor can still see the more accurate color values, and since CC/filters/effects are cumulative and sequential, the difference can be substantial.

[Gopher_Greene]

Aren't most the NLE internally much greater than 8 bit? The new Premiere says it is 32bit internal color processing, so if all I can get in is 8 bit or 10 bit color, it doesn't seem there would be any internal loss. I guess what is confusing me is the actual information of each pixel. Ok, so if I have a single pixel in 4:4:4 then I have three 8bit words of color information for each pixel, correct? Then how are those three 8bit words transfered to the NLE. Are they transfered as three 8 bit words or combined to form a single color word, which is either only 8 or 10 bits long? If so then with three eight bit words we have 32 bit color. If not it seems we're extremly compressing color information.

[Disco Robo]

Don't confuse sample frequency with word size, or think that they are inherently related. 8 bit is 0-255 colors per channel, 10 bit is 0-1023, 12 bit is 0-4095. Sample frequency refers to how often the color is sampled 4:4:4 is every pixel, 4:2:2 is every other pixel, 4:1:1 is every 4th pixel, and 4:2:0 is every other pixel on every other line. You can have any of these combinations interacting with each other the limitation is typically the data rate.

[Gopher_Greene]

Right I'm not confused about that. However 4:4:4 is considered raw RGB no subsampling occurs, So I'm guessing that a white pixel would be three 8 bit words of 255 correct? So once I have these three words how are they processed and sent to the NLE? When FCP says 10 bit color with 4:4:4 uncompressed, does that mean these there 8 bit words are combined to creat a single word 10 bits long?

[Gopher_Greene]

No one has any idea how the color information is transfered from the sensor to the NLE? I guess this is a quest now.

[Arson]

dvcpro is 4:2:2
you can edit that in a 4:2:2 project or a 4:4:4 project

[electricpig]

Right I'm not confused about that. However 4:4:4 is considered raw RGB no subsampling occurs, So I'm guessing that a white pixel would be three 8 bit words of 255 correct? So once I have these three words how are they processed and sent to the NLE? When FCP says 10 bit color with 4:4:4 uncompressed, does that mean these there 8 bit words are combined to creat a single word 10 bits long?
10-bit 4:4:4 is 3 channels (RGB), each with 1024 colours per pixel.
30-bit colour in other worlds like DTP, or billions of colours.

8-bit 4:4:4 is 3 channels (RGB), each with 256 colours per pixel. 24-bit, or millions of colours.

When people talk 32-bit it means 3 x 8-bit, plus an 8-bit alpha channel, or RGBA

[Gopher_Greene]

Thanks for coneecting the dots for me. I read a lot of white papers last night and still couldn't find a clear cut answer.

[BenB]

Also, when you talk about the computer's processing of data, you can say 32-bit, and mean that the software and CPU can calculate the algorythms at a 32-bit word rate, but it has no bearing on the bit depth of the video codec, only on how fast the CPU/Software combined can manipulate the numbers.

"32bit internal color processing" is not the same as saying "32bit color space". Processing is a matter of how fast math can be done to give you the results on screen, not the bit depth of the codec itself.