VV 8k : BAYER? Maths Question

[Antony Newman]

@Red (+Graeme!)

At a sensor resolution of true 8k (8192 x 4320) - would RED reconsider use of the Bayer pattern?

If each 'RGBG' x 4k block moved to 'RGBX' where X was a Dragon Monochrome pixel

- The sensor would still resolve > 4k in all colours (ie a step up from 6k Dragon)
- At High brightness levels, the SNR of the RGB pixels would give images images the existing Dragon 6k look.
- In low light conditions the X pixel would give a 2.5 stops improvement in luma noise
- And (I believe that) for the bottom 15+ stops - the X pixel would also increase colour saturation.


This 8K VV sensor + STH_OLPF + CF_new_processing => Would then have a 3 stop improvement in low light compared to 6K Dragon.

Thoughts?

AJ

 

[Brian Merlen]

doesn't larger pixels help too? would red ever consider going the other way with larger but less pixels on Vista too as a special low light version or something? theres a ton of dif ways to skin a cat...

 

[Antony Newman]

Larger pixels with a RGB filter on the front would improve signal to noise ratio in low light.

At 6k -> the Dragon does not have enough Red or Blue pixels for 4k x 2k image (and so heavily leans on the Bayer to reconstruct colour site in the debayer).
It does have more than enough Green pixels for a 4k finish.

At 8k -> if the VV sensor uses a Bayer pattern -> it will have Exactly the right number of Blue and Red pixels for a 4k finish.
It will also have TWO x Times the number of green pixels than is needed for a 4k finish.

I am suggesting that at Some resolution, a Bayer pattern is no longer logical to use - and that that point may well be now (as we have enough resolution in Green).

I speculate it is time to exchange half the Green pixels with Clear Monochrome pixels ... to gain the Latitude, Contrast, Low noise, and Sensitivity of a Monochrome Dragon pixel.
If the VV sensor had 4k x 2k monochrome -> it would inherit the 'die for' Monochrome qualities in LOW light conditions - but now with the Colour purity of the Colour dragon sensor when there is light available.

Where starved of light sensor points cause fluctuations in colour brightness - this extra pixel would improve colour linearity by 2.5 stops.

It would then be 'safe' to use more low level data as the DR range of camera would allow the Useful range to extend into Monochrome territory.

This isn't magic -> this is an exchange of Green resolution for Usable dynamic range.

There is a Math part involved here (bayer / debayer ) -> But there some other hidden benefits ... like colour linearity from the X pixel up until it clips.
This could be used to fix the 'kinks' that currently exist in the Dragon sensor's mid to upper linearity (compared to the larger sites in say the Alexa photosites).

AJ

 

[Antony Newman]

Quick search on the net reveals that there are other companies trying to improve low light colour rendition for smart phone cameras.

LINX:

http://www.scribd.com/doc/261875793/LinX-Imaging-Presentation

====

Another way to look at this was ... let say that Red could release a 16k x 8k sensor that had Dragons colour performance .... or in the same area a 12k x 6k sensor that had 2.5 stops more usable dynamic range at the bottom end ... You'd jump on the lower resolution sensor .. as we can use the 2.5 stops today.

If the existing 6k Dragon sensor resolved to 3.6k (recent test) -> and the 8k sensor (with X) pixel improved this to >= 4k .... what would you want?

Thoughts?

AJ

 

[Craig Parkes]

Given that there is probably a lot of literature on converting the straight Luma value back to green due to Y'Cb'Cr' encoding, this sort of modification to the Bayer pattern does seem to make a lot of sense if it can give increased sensitivity to be used for processing out noise (Kind of like ADD but not requiring multiple frames.)

I wonder if such a methodology exists, at least in a patent form.

 

[alexjentz]

... very interesting thoughts!

 

[Dr. Sassi]

Brainstorming:

You would have the need to read out the X pixel shorter, e.g., 30º shutter vs 180º, as the "charge" of the X would fill faster. With that the motion blur will change, but only for that photo-site, otherwise you have clipping. Perhaps read it out twice... which seems impractical.

If you (however) get this done, just for the argument sake, the different DR results in a larger problem with the AntiAliasing calculation, as you can see it often in 3D render applications with "hot" edge-values.

Reading out the X differently would disable the Motion Mount to a certain degree (hence no HDRx while MM is mounted.)

Currently the double green is certainly welcomed for VFX. If you take that one green away, the green-screen resolution might be only half.

(A decade or so the Sony 828 had an fourth pixel in a special green, and the color-science were extremely difficult; hence Sony stopped it. Other tried with yellow after that fir screens, but not I haven't seen it in a high end camera )

Perhaps RGBL ... Luminance would be more communicative.

/brainstorm