4:4:4 Color Decimation

[ddixon]

Does SCARLET support 4:4:4 color decimation?

 

[Peter Majtan]

Does SCARLET support 4:4:4 color decimation?

If by 4:4:4 You are referring to RGB mode - apparently it does.
4:4:4 YCbCr - not sure...

 

[Stephen Gentle]

Does SCARLET support 4:4:4 color decimation?

There is no 4:4:4 on Scarlet.

Scarlet will record whatever comes off the sensor. My understanding is that a single colour channel (bayer pattern filtered) comes off the sensor, which is wavelet-compressed and dumped onto CF. Later, the image is de-compressed (with wavelets you can choose to decompress the full image, or a fraction of the size), re-constructed, and you get a full RGB image.

So if you're asking whether there is no chroma subsampling, then the answer is yes, but from what I've heard, I don't think that x:x:x applies to REDCODE at all.

I'm sure Graeme can explain it better.

 

[Peter Majtan]

There is no 4:4:4 on Scarlet.

Scarlet will record whatever comes off the sensor. My understanding is that a single colour channel (bayer pattern filtered) comes off the sensor, which is wavelet-compressed and dumped onto CF. Later, the image is de-compressed (with wavelets you can choose to decompress the full image, or a fraction of the size), re-constructed, and you get a full RGB image.

So if you're asking whether there is no chroma subsampling, then the answer is yes, but from what I've heard, I don't think that x:x:x applies to REDCODE at all.

I'm sure Graeme can explain it better.

"UP TO 100 MB/SEC REDCODE RAW AND RGB RECORDING TO DUAL COMPACT FLASH"
from the official Scarlet's specs. It will record directly RGB footage - it is one of the more controversial features (as Red One is not at present capable of pulling this trick)...

 

[Barry Green]

Guys, you can't get 4:4:4 from a single sensor.

You could record whatever you get AS 4:4:4, but that doesn't mean you're getting the level of color detail that true 4:4:4 would mean. It's just not possible. You have to have three chips to get true 4:4:4.

 

[Peter Majtan]

Guys, you can't get 4:4:4 from a single sensor.

You could record whatever you get AS 4:4:4, but that doesn't mean you're getting the level of color detail that true 4:4:4 would mean. It's just not possible. You have to have three chips to get true 4:4:4.

Then how do You get true RGB (even at 8:8:8 - just bothering Graeme :innocent: ) from a single CMOS sensor SLR?

(This is assuming that the 4:4:4 is referring to RGB of course)

 

[Lucas Wilson]

"UP TO 100 MB/SEC REDCODE RAW AND RGB RECORDING TO DUAL COMPACT FLASH"
from the official Scarlet's specs. It will record directly RGB footage - it is one of the more controversial features (as Red One is not at present capable of pulling this trick)...

Ok... first of all...

Next person to use x:x:x nomenclature when referring to a CMOS sensor gets smacked upside their technical wazoo.

Second...

A CMOS sensor can't directly record RGB. Just ain't in the nature of how a CMOS sensor is put together. Perhaps RGB will be "assembled" within the camera circuity, but I doubt that Scarlet directly records RGB footage.

Unless there are some rules of Physics I ain't understanding. I will open up to being smacked by Graeme or Deanan if I'm missing something.

Lucas
-----
Reader of Patents
ASSIMILATE, inc.
LA, CA, USA

 

[Peter Majtan]

Ok... first of all...

Next person to use x:x:x nomenclature when referring to a CMOS sensor gets smacked upside their technical wazoo.

Second...

A CMOS sensor can't directly record RGB. Just ain't in the nature of how a CMOS sensor is put together. Perhaps RGB will be "assembled" within the camera circuity, but I doubt that Scarlet directly records RGB footage.

Unless there are some rules of Physics I ain't understanding. I will open up to being smacked by Graeme or Deanan if I'm missing something.

Lucas
-----
Reader of Patents
ASSIMILATE, inc.
LA, CA, USA

My bottom hurts already (I was just teasing Graeme - we had quite a thread regarding this little back)...

But that does not answer my question:

How does a single CMOS sensor inside a DSLR (still) capture RGB?

 

[Chris Kenny]

But that does not answer my question:

How does a single CMOS sensor inside a DSLR (still) capture RGB?

If you're having the camera produce an RGB file like a TIFF, rather than a raw file, then the file format supports full-resolution color data. But if you actually point the camera at a resolution chart and examine one color channel by itself, you'll see it's a bit soft. This is because the sensor isn't actually resolving full resolution chroma. (It isn't quite resolving full resolution luma either, but it's closer.)

Of course, Scarlet's RGB recording mode will probably be 1080p, not 3K, so the downscaling will get you closer to having chroma and luma resolution that matches the pixel count of the recorded file.

 

[Lucas Wilson]

If you're having the camera produce an RGB file like a TIFF, rather than a raw file, then the file format supports full-resolution color data. But if you actually point the camera at a resolution chart and examine one color channel by itself, you'll see it's a bit soft. This is because the sensor isn't actually resolving full resolution chroma. (It isn't quite resolving full resolution luma either, but it's closer.)

Of course, Scarlet's RGB recording mode will probably be 1080p, not 3K, so the downscaling will get you closer to having chroma and luma resolution that matches the pixel count of the recorded file.

Hey Chris,

I think the point is that a CMOS sensor doesn't record RGB period. There is always a conversion step somewhere.

Lucas

 

[Andrew M.]

You can convert anything to anything if you have enough CPU power in the camera, except US to Euros at 1:1 ratio.
CMOS is single channel Bayer not RGB. Some camera shoot straight in to jpg if you wish so.

 

[conrad gaunt]

Beyer sensors are basically high quality grey scale images, with one filter (either R,G or B) over each sensor pixel. Each sensor pixel is missing two colour channels, which are then interpolated from surrounding sensor pixels during the de-mosaic process. De-mosaic processing can be simple (for realtime preview etc) or complex in nature (done in post). Either way, the missing colour channel have to be calculated/guessed at. In practice, I`d rather use a 3k CMOS camera than a 1080p 3chip system, as you will get visually better results, especially if an efficient wavelet compression is being used instead of a DCT based MPEG compression or similar, and you do a HQ debeyer. There`s no real point in recording RGB data from a beyer captured image, as you simply triple the amount of data you need to compress, which you can generate later from the original beyer data anyway.

 

[Peter Majtan]

You can convert anything to anything if you have enough CPU power in the camera, except US to Euros at 1:1 ratio.
CMOS is single channel Bayer not RGB. Some camera shoot straight in to jpg if you wish so.

Comparing Bayer with RGB is very misleading as Bayer's photosites consists of R, G and B elements (with double G's then R and B respectively)...

So whatever processing is going on inside the camera the original RAW sensor data are RGB values...

 

[lion]

Hi All,

First post, long time lurker. Now with Scarlet in my price range, figured should start to learn about the products more in depth.

Does anyone have any good links or white papers that breaks down how color is processed in RED cameras and the associated RED work flow?

I'm curious to learn about the technical side in a general article. Perhaps something like the following link on sensors.

http://www.digiphoto.org.uk/sensors.htm

Like everyone else really excited about Scarlet. Can't wait till it's available.

Thanks for your help!

 

[Graeme Nattress]

Current plan, as noted in the brochure, is RAW and RGB recording. As for the form the RGB record will take, I don't know yet.

Barry, you're confusing the issue. And of course, quite frankly, you can't get 4:4:4 from a 3 chip system either (if it's got any offset in the chips) and if the chips are perfectly aligned but you have the correct OLPF you don't get 100% resolution in any of the channels. Or you can have 100% resolution, but objectionable aliasing. The only camera that could really achieve 4:4:4 at it's native resolution is one with the Foveon, but then the chroma is so noisy they use aggressive spatial chroma noise reduction as a standard part of their decode, so it's not 4:4:4 either, and it aliases nasty as they "forget" to put in an OLPF.

Graeme

 

[Graeme Nattress]

http://www.digiphoto.org.uk/sensors.htm

Problem with that article, is like every diagram showing how a Foveon sensor works, it colours the layers red, green and blue, but that's not the colours you get out! With a Bayer sensor, even without colour matrix, the colours are "close" to reality and quite saturated. RAW from the Foveon is very unsaturated and uncolourful. The luma is clean, but the chroma quite noisy, necessitating strong spatial chroma NR as part of the RAW processing.

And, of course, the Sigma cameras omit the OLPF so alias quite badly, which is bad enough on a stills camera, but would be totally awful on movement in a movie camera. Of course, the vast majority of Sigma users seem to like aliasing, but I don't, and for very good reason.

Graeme

 

[Dory Breaux]

Regardless of any numbers or arguable specs, the Ones pictures are BEAUTIFUL, and we should expect no less from scarlet.

 

[Graeme Nattress]

Dory, you hit it in one!

Graeme

 

[lion]

Problem with that article, is like every diagram showing how a Foveon sensor works, it colours the layers red, green and blue, but that's not the colours you get out! With a Bayer sensor, even without colour matrix, the colours are "close" to reality and quite saturated. RAW from the Foveon is very unsaturated and uncolourful. The luma is clean, but the chroma quite noisy, necessitating strong spatial chroma NR as part of the RAW processing.

And, of course, the Sigma cameras omit the OLPF so alias quite badly, which is bad enough on a stills camera, but would be totally awful on movement in a movie camera. Of course, the vast majority of Sigma users seem to like aliasing, but I don't, and for very good reason.

Graeme

Thank you Graeme for the thoughtful and quick response to a total noobie question. Color is an area which is completely new to me and as such a good avenue for me to learn correctly what RED does without any preconceptions on my part.

Do you have books or suggested links to read how the RED sensor works thru the work flow process?

Now, I wonder if RED has it's manual online for PDF download, so I can start doing some reading over the rest of the year.

EDIT: For any other newbies, look under support for RED One downloaded pdf. I think I resisted looking before because the simple beautiful functionality makes me want one too much!

Way cool, you download REDCINE and start playing with some sample files.

 

[Graeme Nattress]

We're understandably reticent to go into details of the nuances of how Mysterium and the image processing pipeline and REDCODE RAW works. Generally, when I write on the forums, I'm talking about image processing in general, but obviously with direct application to what you and we do with the RED footage. For instance, I'll talk about demosaic algorithms in general, I'll talk about the quality of ours, but not the specifics of how it works, as that is proprietary.

I hope to find some time to put some white papers on various aspects of RED technology together, as I truly believe that a educated camera user is a better camera user.

Graeme