Jump to content

Is 422 from a 4k sensor real 4k 422?


Recommended Posts

I've been told the information in a 4k image from a 4k sensor is anything from 33% to 70% on a bayer sensor. I am so confused with so much misinformation around. I've looked through the forums and learned a lot about this topic. But it never answered this question.

 

 

- Is 422 from a 4k sensor real 4k 422?

- Exactly how much real color information is in a 4k sensor at 4k resolution (pixels)?

 

You get the jagged edges when trying to key a 420 image... Why would a 4k 422 sampling from a 4k sensor be any different. If its not enough information there will it look like 420 even at 422?

Link to comment
Share on other sites

  • Premium Member

You're mixing up a few issues here...

 

In terms of resolution, there are two things here: the pixel resolution (just a count of the pixels in the file) and the measured resolution (determined by shooting charts).

 

A bayer pattern has 50% of the photosites filtered green and 25% filtered red and blue each.

 

But that doesn't mean that for a 4K sensor, you end up with a max resolution of 2K in the green channel and 1K in the red and blue channel each, it's not quite that bad.

 

Each color channel is processed back into 4K to create 4K RGB from 4K raw. The overall image, when measured by shooting charts, ends up somewhere around 75% of the pixel resolution of the sensor (on a good day.)

 

Now imagine that you took off the bayer filter to create a monochrome sensor. Even then, it would be hard to achieve 4K measured resolution with a 4K sensor because there is an optical low-pass filter to reduce aliasing. Now maybe if you removed that as well, you could achieve 4K resolution on a chart but you'd see a lot of moire artifacts because the Nyquist Theorem states that a sampling rate has to be 2X the frequency of the original to avoid aliasing (in other words, you'd need an 8K sensor to get a 4K image with no moire at all if you didn't want to use a low-pass filter on the sensor).

 

So to sum it up, you shoot on a 4K bayer sensor and you can end up with a 4K raw recording processed into 4K RGB files, but the measured resolution (from shooting line resolution charts) in the image will be more like 3.5K, let's say. But there are a lot of factors that might lower this number further (a soft lens, bad focus, a diffusion filter, a hazy room, etc.)

 

And if you measured those charts separately in the green, red, and blue channels, you'd probably find that the red and blue channels are a little softer.

 

4:2:2 is a color subsampling value -- you can convert a raw recording to 4:4:4, 4:2:2, 4:2:0, etc. 4:4:4 means that each color channel is given an equal amount of bandwidth. 4:2:2 means that red & blue have half the allotted bandwidth compared to luminance/green (one argument for shooting green screen instead of blue screen... but a bigger argument for working with 4:4:4 if you are doing chroma keys.)

 

4:2:0 is, I believe, worse than 4:2:2, somewhere in between 4:2:2 and 4:1:1... you can look it up.

 

So in a sense, shooting with a bayer sensor camera means that your red and blue information are a little softer than your green information but using a 4:2:2 recording instead of a 4:4:4 recording means that your red and blue channels are also now more compressed, so even softer than they are in the 4:4:4 version.

 

Keep in mind that even if you scan 35mm film at 4K per color channel for a "true" 4K RGB recording, the color layers in film stock are not equally sharp either.

  • Like 1
Link to comment
Share on other sites

  • Premium Member

I forgot to add that there are other quality issues that will affect your ability to pull keys - not all 4K 4:2:2 recordings are the same. There is the issue of bit-depth (8-bit, 10-bit, 12-bit), overall compression, and whether that compression scheme uses inter-frame or intra-frame compression schemes, etc.

Link to comment
Share on other sites

You are the specific person I was learning from through these forums. You had a lot of valuable information I couldn't find anywhere else. I am new to forums so I don't know how to quote properly yet.

 

"A bayer pattern has 50% of the photosites filtered green and 25% filtered red and blue each.

But that doesn't mean that for a 4K sensor, you end up with a max resolution of 2K in the green channel and 1K in the red and blue channel each, it's not quite that bad."

Great this is what I needed to know. So in other words ,if indeed the most information you could possible get out of a bayer sensor is 75%, that means the 4k 422 is indeed what it says and not fake pixels. That is if you have enough information depending on how you shoot it right? Even then 35mm is not always accurate. I need to see some examples of test charts. Trying to learn as much as possible about this subject.

Link to comment
Share on other sites

  • Premium Member

Again, there are other factors that will affect quality -- you might find it easier to color-correct and do keys with 12-bit 2K 444 ProRes XQ from an Alexa versus some camera that records in a very compressed 4K 4:2:2 format, particularly if it is only 8-bit.

 

Not sure what makes a pixel fake or not. You have to keep in mind that the conversion from a raw bayer-pattern signal to color is a transformation, it's not like the photosites off of the sensor are just turned directly into pixels, you've converted one system to another, plus there is an analog to digital conversion in there as well, plus the recording codec, so I'm not sure what a "true" versus a "fake" pixel would be.

 

All that 4:2:2 means is that the luma gets sampled at 4X and red-luma and blue-luma are sample at 2X. Basically chroma information has half the bandwidth compared to luminance.

 

But certainly a lot of people have asked this question of 4K 4:2:2 versus 2K 4:4:4, which is better. Sort of depends too and where you have to end up, with a 2K master or a 4K master? For some vfx work, they'd prefer the 4K picture because it gives them more flexibility to reframe if ending up at 2K.

Link to comment
Share on other sites

Another wrench to throw in the pot:

 

If you shoot a 4k format that records 4:2:0 or 4:2:2, you can down-sample the 4k to 2k and get a 4:4:4 sample. It won't affect the bit-depth any, but doing that would certainly help with grading and chroma key work. Doing a 4x down-sample would also act as a built-in noise reducer, and possibly hide some of the compression artifacts.

 

Just throwing it out there.

Edited by Landon D. Parks
Link to comment
Share on other sites

Another wrench to throw in the pot:

 

If you shoot a 4k format that records 4:2:0 or 4:2:2, you can down-sample the 4k to 2k and get a 4:4:4 sample. It won't affect the bit-depth any, but doing that would certainly help with grading and chroma key work. Doing a 4x down-sample would also act as a built-in noise reducer, and possibly hide some of the compression artifacts.

 

Just throwing it out there.

 

 

 

Yeah I've heard of this. Although I don't know the technical side of it.

Link to comment
Share on other sites

4:2:2 is a color subsampling value -- you can convert a raw recording to 4:4:4, 4:2:2, 4:2:0, etc. 4:4:4 means that each color channel is given an equal amount of bandwidth. 4:2:2 means that red & blue have half the allotted bandwidth compared to luminance/green (one argument for shooting green screen instead of blue screen... but a bigger argument for working with 4:4:4 if you are doing chroma keys.)

 

This goes against my understanding of how chroma sub sampling works. It is my understanding that in order to use subsampling the RGB data is first converted to YUV colorspace. In this colorspace Y is luminance or the brightness of the pixel and UV are a 2D coordinate that defines the color of the pixel. Chroma subsampled values are then described in the format Y:U:V so 4:2:2 means that for every 4 pixels in a row 4 pixels had Y/ Luminance, 2 had U and 2 had V so basically all of the colors red, green and blue are allotted half the bandwidth of the luminance channel. I am just wondering if I have been mistaken.

Link to comment
Share on other sites

  • Premium Member

I guess you are correct... though I believe green information is contained in both of the two chroma channels, so it may end up with better resolution than red or blue. Or maybe green information has more resolution simply because it has twice as many photosites in the bayer pattern. Or maybe luminance is mainly derived from the green channel so resolution of green information is contained both in the "Y" but also the "Cb" and "Cr"... Phil Rhodes can answer this better than me.

Link to comment
Share on other sites

Something else just came to mind. I'll try to explain this the best way I can. There are 3 question and I can finally move on from this.

 

4K sensor

Green is 50% = 2k

Red is 25% =1k

Blue is 25 = 1k

 

Thats 3k (75%) worth of color information. So the red and blue channel is less than half the resolution of 4k.

 

1. Wouldn't that make it 4:1:1? Im not counting the fact that film may not have "true"colors or the subject being out of focus etc. Im talking about strictly whats going on inside the camera.

 

2. If the first question is true, then the 4k 422 from a camera with a 4k sensor is not the actual information from the sensor. (Given that it does not need to be, for learning purposes I need to understand this. Im not trying to get this magical perfect quality, Its just that if I had to explain this to someone, I'd like to be able to.)

 

3. Also if this is true, will the 411 give me the benefits of 422 raw to proves 422? Or is it just a 411 in a 422 shell and will behave like 411?

 

I'll write checks for helping me lol. Maybe not...

Link to comment
Share on other sites

  • Premium Member

Raw is not 4:2:2 because 4:2:2 is a color subsampling scheme and raw isn't a color image yet. Raw can be converted to 4:4:4, 4:2:2, 4:1:1. There is no such thing as "422 raw".

 

No a bayer sensor isn't 4:1:1. It's not 4:2:2 either. It's not 4-anything!

 

What do you mean by "not true information from the sensor"? This sounds like you are talking about "fake pixels" again.

 

Color subsampling is a compression scheme basically where the color is more compressed than the luminance. That's not how a sensor works. A sensor is an ANALOG device, signals from the photosites have to be digitized and at some point, the bayer pattern in the raw information has to be de-mosaiced into a color image. THEN, it can be color subsampled. Or not. Like I said, it's a transformative process.

 

Only the crudest debayering algorithm simply takes the 50% green photosites and just converts them to the green channel information and then upscales it back to 4K. See:

 

https://en.wikipedia.org/wiki/Demosaicing

 

No, there are no benefits to 4:1:1 except that it takes up less data.

 

We've got one thread going here where we are debating what ProRes 4444 XQ is, which is created from a raw signal from the Alexa, which uses a bayer pattern sensor -- no one argues that recording 4:2:2 would be just as good as raw or 4:4:4, and certainly no one would argue that 4:1:1 is just as good as 4:2:2.

 

I'm getting the impression that you're trying to justify something to yourself.

Link to comment
Share on other sites

  • Premium Member

http://www.red.com/learn/red-101/bayer-sensor-strategy

http://www.reduser.net/forum/archive/index.php/t-71931.html

 

Another factor if you are really talking about improving resolution in the red and blue channels for eventual conversion from raw to 4:4:4 is how much oversampling in general a sensor is doing.

 

There's a point where the reductive reasoning that states that a bayer signal is "4:2:2" has no real world practical application since there are so many mitigating factors that affect quality, from the actual sensor being used, to the quality of the raw signal, its bit-depth, its compression scheme, and even if recorded as 4:2:2, the particular recording codec used.

Link to comment
Share on other sites

My head is going to explode lol. I feel like an idiot because I have been told a bunch of stuff in Facebook groups to go off of. Im trying to forget what I learned. I was just looking for a clean way to explain the 25% of information from the red and blue channel using the info you told me above. I am no expert, Im not trying to justify anything by any means. Im just a guy trying to learn.

 

I know raw isn't 444 422 420 etc, I was just going by the information that was there before the raw is processed. Maybe I should have used the term "resolution" of photosites?. I thought it was clean cut to convert it to properly based on the information there. I was wrong. I will throw everything out I just said. I want to start over completely!

 

So according to the links you gave me and what you have said... I will sum things up for myself.

 

From the wikipedia

 

  • To reconstruct a full color image from the data collected by the color filtering array, a form of interpolation is needed to fill in the blanks. The mathematics here is subject to individual implementation, and is called demosaicing. The reconstructed image is typically accurate in uniform-colored areas, but has a loss of resolution (detail and sharpness). Digital camera typically has means to reconstruct a whole RGB image.

 

So to sum that up, the sensor creates a full RGB image like you said. During the transformation, It can be sampled to what ever you like, but it somewhat lacks the quality or color accuracy of a RGB image from a 3CCD type camera? Image interpolation has it faults but at the sometime very good. In most cases shooting with these cameras you'd hardly notice since our eyes aren't as sensitive to it. I believe this is the reason you told me about 35mm not being perfect. Things can be out of focus among a lot of factors restricting resolution.

 

Please tell me this is right Haha. If I am wrong I give!

 

 

If anybody else is reading this trying to learn with me, this video helped me out too.

Edited by Kendrick Gray
Link to comment
Share on other sites

  • Premium Member

Your description is accurate.

 

The situation these days is that modern cameras have so much excess resolution that seeing sharp edges on brightly-coloured objects isn't really that big a problem anymore. Previously, if you shot, say, a brightly-coloured blue flower against a background of, say, red flowers, it was possible to see the blue and the red blurring into one another as the camera couldn't quite figure out where the edge of the colour was supposed to be. Now, if you're looking at an HD picture derived from a 4K camera, the error more or less gets lost as the image is scaled down.

 

The problem that can persist is one of colour noise. For the Bayer reconstruction to work well, the red, green and blue filters need to be reasonably unsaturated - that is, fairly pale colours. This is so the red, green and blue pixels can each see something of the whole scene, to guide the reconstruction. This works OK, but would ordinarily create a rather desaturated picture with muted colours. To correct that, the reconstruction process more or less just turns up the saturation, just as you would in Photoshop. Try it, and watch the picture go noisy in a particularly specific and unpleasant way. Bayer reconstruction causes noise as well as softness and that, if anything, is the problem with it in current practice.

 

Three-chip cameras are now rare, though, so this is largely a theoretical discussion. Enormous improvements in the performance of all these things have happened in the last ten years, and the results are now very good regardless.

 

P

Link to comment
Share on other sites

  • Premium Member

Yes, obviously a 3-sensor camera is going to have relatively better resolution of the red and blue compared to the green, but again, there aren't any 4K 3-sensor cameras out there, at best there are 2/3" 1080P 3-sensor cameras.

 

And there are also the RGB striped single sensor cameras -- the Panavision Genesis and the Sony F35 -- which made an excellent 1080P color image because there were 2MP of photosites each for red, green, and blue. But having the stripes side-by-side so that the next green stripe was separated by two stripes from the last green stripe, etc. did cause certain issues with chroma moire.

 

This is where the higher megapixel CMOS bayer-filtered sensors have some advantages from oversampling if you are going to end up with a 2K finished image, let's say (or if you are going to end up in 4K but start out with a 6K or 8K sensor.)

 

The logic that a bayer-filtered sensor is inherently "4:2:2" is not completely bogus in the sense that both operate on the principle that human vision is more sensitive to resolution in luminance/green than in red and blue but that's still such a crude analogy as to do more damage than good since 4:2:2 is really about giving less data for color over luminance, and data does not exactly equal resolution or color information, some compression techniques are more lossy or lossless than others, and the actual subject may or may not have problems with those schemes.

Link to comment
Share on other sites

Yes finally! Yeah I figured it wasn't any 4k ccd type cameras yet. I noticed that the c300 and c500 do 2k RGB at 444. Thanks David Mullen for sticking with me on this. Thanks for your input also Phil Rhodes and the others! I can sleep at night and not have to worry about having nightmares about this.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...