confused.. Is RED one's color sampling 4:2:2?

[Beom Koh]

I was reading a magazine today about Video Camera and it was telling me that RED ONE's color sampling is 4:2:2..

I thought since RED uses RAW, it could output to 4:4:4 if I want..

Am I wrong here?

 

[Gunleik Groven]

You're not wrong, but the answer is complicated. RED is recording raw, not x:x:x subsampled RGB images.

Graeme will have to do the correct explanation and usually he is all over these questions.
I'll add a question:

If I have understood this right-ish, you CAN indeed get a "proper" 444 RGB 2k/1080, maybe even 3k 444 image out of the 4k recording, but at 4k the sampling would practically be...?

Cheers!

Gunleik

 

[Graeme Nattress]

We don't use any chroma subsampling techniques. The only place where 4:2:2 is used is for the HD-SDI (to adhere to it's standards).

When people refer to 4:2:2 they often use it incorrectly....

Graeme

 

[Gunleik Groven]

LOL... I'll delete this post if improper, but when you go to RGB/YUV a 3 channel image is eventually supplied. Wouldn't it then be right to thhink that there is a corelation between the downsampling and the chroma (not luma) precission?

 

[Graeme Nattress]

4:2:2 are not measures of chroma precision compared to luma. They're a description of a specific chroma sub-sampling process that some cameras employ. If you don't use that process, that terminology is invalid.

Graeme

 

[Gunleik Groven]

OK... So saying that I make a 444 1080/2k cineon compliant DPX for online is incorrect?

I guess not (but I am no engineer...)

Question would be, what would the 4k DPX be?

Anyhow. This is just a thing I am curios about, and has no practical implications.

 

[Graeme Nattress]

Does dpx format even allow for chroma sub-sampling? If so, I've never come across one! 4:4:4 is like saying no chroma sub-sampling has occurred, so while not technically wrong to say it, it's rather a redundant statement.

Graeme

 

[Gunleik Groven]

THX!

 

[Corrado Silveri]

Thanks Graeme,
your answers always spread some light to my limited knowledge...

And kudos to Gunleik, you always get the right question

 

[Stacey Spears]

Does dpx format even allow for chroma sub-sampling? If so, I've never come across one!

I had to deal with a 720p60 10-bit 4:2:2 DPX sequence last year from Fotokem. Very few products could handle 4:2:2 DPX. Even fewer could handle what the DVS was producing. After some debugging, it turned out that the DVS was outputting DPX in big endian when everyone was expecting little endian. Moral of the story, use DPX v2. (SMPTE 268M-2003)

 

[Graeme Nattress]

Wow - seems DPX does have a spec for 4:2:2 thus:

100 CB, Y, CR, Y (4:2:2) -- based on SMPTE 125M
101 CB, Y, a, CR, Y, a (4:2:2:4)
102 CB, Y, CR (4:4:4)
103 CB, Y, CR, a (4:4:4:4)

I must admit I'd never previously heard anyone actually use it.

Graeme

 

[RivaiC]

We have encountered this DPX YUV format a few times. A high speed mode from Arri that records the file with third party OB1 hardware. You get DPX YUV. Quality isnt great

 

[Dan Hudgins]

Bayer vs. your eyes

Bayer vs. your eyes

If you de-Bayer and R3D that is a bit underexposed and then look at the RGB seperations you should see some difference in the "resolution" of bright colored parts of the images. That is beacuse the Red and Blue are 25% pixel fill and the Green is 50% pixel fill. Because of the sharpen to compensate for the OLPF filter needed, the interpolated images resolution is a bit more complex to measure because all three sensor colors pick up all three colors.

Monochrome test charts may show the red and blue with a bit more aliasing and noise, but that depends on the sharpen used, and RED tends to adjust the sharpen so that aliasing is under control making all three images a bit soft, but you may see the green image showing a bit more detail.

This is not a big issue, since your eyes cannot see much small detail between green and blue, and are only sonewhat better on strong red colored subjects. The NTSC color system was based on the lack of visable detail at a given distance from the screen on strong colors.

The luma detail from the green image and some from the mixing of the red and blue, are superimposed over the colors, so you do see detail in colored areas, its just fine changes in hue that you cannot see. Most of those small hue changes are noise in the sensor signal, since the negative matrix makes the chroma noise much worse unless you low pass it or at least do not compensate for the OLPF blur in the chroma.

It seems odd to me that so many people are looking for a one sentence lable for everything, the world is not as simple as what you can put into one sentence, except maybe for saying that Nixon was a crook...

 

[paulherrin]

I don't know of any sharpening happening in any RED camera. That's always been left for post, if desired.

 

[Dan Hudgins]

Before or after compression

Before or after compression

I don't know of any sharpening happening in any RED camera. That's always been left for post, if desired.

Acutally if you think about it, it would be better to do the anti-OLPF blur before the Wavelet compression, than after, since sharpening after in post could bring up any wavelet artifacts.

Its possable that there is a sharpen in the camera for the viewfinder circuits to make it easer to focus the camera to compensate for the OLPF blur?

If you have seen image data that has not been compensated for the OLPF blur it can be quite soft, though RED has an advantage goint out to a 1280x720 monitor from a 4.5K image since that reduction would require some blur before the downsize anyway to reduce aliasing. At 1:1 though, the image would be quite soft without some peaking.

An interesting point about doing any sharpen in post is that it sharpens BOTH the luma and chroma, unless you are using some special image processing, where as if you use the anti-OLPF blur compensation in the de-Bayer it should JUST sharpen the luma and not the chroma, since as I pointed out the chroma noise is higher because of the color matrix, its not that the red and blue have zero bandwidth on the pixel level like in sub-sampling, but you don't want the same MTF curve on the chroma as luma for results that look good from a normal viewing distance, where you eyes don't notice the lower MTF in the chroma much, but may see less chroma moire and aliasing and noise.

 

[paulherrin]

hmm interesting. have you had any results experimenting with sharpening the luma channel only (in post)? would you recommend that as a practice or something to try out, or are there and curious limitations and pitfalls to that technique? i know i've done the opposite with slightly blurring 4:2:2 chroma pixels as part of a process to get rid of jaggies, especially for vfx application, which would make the luma slightly more sharp, relatively. but i've never just sharpened the luma, i guess i'll have to try it then.

 

[Dan Hudgins]

luma vs. chroma

luma vs. chroma

Yes, in the new de-Bayer and image processing program I am working on, I duplicated my de-Bayer processing AFTER the interpolation step for doing RGB files (BMP 24bpp, CIN 10bpc, DPX 10bpc and 16bpc, TIF 16bpc), with processing of RED (tm) footage in mind.

There are several filters in that process method, so you can load a 16bpc DPX or TIF made by REDCINE-X (tm) and color correct along with area filter and sharpen the luma, and that has options of active or passive sharpen to control noise build up.

One of the problems with my workflow now for film scans is that if you matrix the color, the film grain goes up because the chroma is not well defined for single pixels if the film scan is very sharp. With the new program the grain should be better controlled since the chroma filter radius is adjustable. The same goes for sensor noise, the chroma for single pixels has more noise that the average of a small patch.

Anyway, as far a RED (tm) footage goes, when I get my new program out it should be able to batch process RED (tm) footage with better results and higher speeds (well unless you use additional filters) since its a 32bit Windows Console program (also 32bit DOS and GUI versions, 80x25 interface in all three, 32bit DOS graphics preview works up to 2048x1536, and Console up to 1600x1200 so far.).

I'm working on the documentation now and hope to have it posted for "Beta Testing" before too long.

 

[mansh singh]

I am confused with all the answers from this forum. So all a nerd knows is 4:4:4 chroma subsampling means ultra quality while 4:2:2 means some what less quality which is not apparently visible by eye but if it is a green screen shoot it matters. So my question is if i shoot in red one . And i used r3d files directly in premiere or after effects , so what chroma subsampling i would be working on

 

[Dan Hudgins]

Green screen

Green screen

I am confused with all the answers from this forum. So all a nerd knows is 4:4:4 chroma subsampling means ultra quality while 4:2:2 means some what less quality which is not apparently visible by eye but if it is a green screen shoot it matters. So my question is if i shoot in red one . And i used r3d files directly in premiere or after effects , so what chroma subsampling i would be working on

Bayer filters have 50% pixel fill for Green and 25% pixel fill for red and blue, that is one reason why video uses Green Screen, rather than the older Blue screen that was used with film cameras.

Chroma sub-sample is not really what RED (tm) would seem to use, it stores the images in R3D files as compressed RGB desaturated signals it seems, so the ratio of RGB information would mimic the Bayer filter. You should be looking into Bayer filter cameras. There do not seem to be any 4K 3-chip cameras now, so there is no 1:1 Bayer to 3-Chip choice anyway. Also, low passing the chroma can imporve the green screen matte edge by reducing the noise, that is not the same as "sub-sample" which would increase aliasing.

In other words, what other choice do you have now in a 5K to 4K camera?

Go see Kaboom (2010) By Gregg Araki, he was in SF and I talked to him about the shooting of his film on RED ONE and he said the reason he used it over shooting on 35mm film was BECAUSE of his need to have many green screen shots, so you can see his results for yourself without needing over simplistic catch phrases, 4:2:2 and 4:4:4 are not how video cameras work today as complex filters and lossy compression with motion estimation and such are used in addition any simple ratios. You will have to search for a town near you and see when a print comes close to where you are, which I cannot tell because you did not put it in your ID yet(?)

http://www.imdb.com/title/tt1523483/

Maybe you can contact Gregg and ask him if he thinks there were any problems with RED ONE for Green screen, as far as I could make out he liked the results.

 

[Gunleik Groven]

Also worth mentioning is:
This is what is done in most still cameras. For some reason we don't have that discussion on the still side...