2 Stop faster than bayer cmos/ccd pattern

[Michael Brennan]

http://www.pcworld.com/article/id,132865-pg,1/article.html
Perhaps a version of RED with this novel filter array just announced from Kodak can be used on the Mysterium?
Kodak are claiming a two stop increase in sensitivity compared to bayer.

Basically they have a fourth pixel that has no filter and this is used for luminance. Trade off is that there are fewer colour samples.

Graeme N, can the RED in camera processing be tweaked to work with different filter array or does the in camera processing go hand in hand with RED RAW?

This idea could also have a use on 3 chip TV camera imagers one chip clear, one green the third with a Red and Blue array.

Mike Brennan

[Stephen Gentle]

I don't know... Wouldn't this knock a lot of data off the colour channels?

People are complaining enough about the resolution of Bayer sensors...

[oldphart]

I don't know... Wouldn't this knock a lot of data off the colour channels?

People are complaining enough about the resolution of Bayer sensors...

Should not make much difference to the resolution, given enough processing power. Colour resolution will be lower, but not worse than what we already see with 4:2:2 I would guess.

I was never much impressed with the Bayer pattern, I think it is a relic from when we did not have fast enough signal processors. While I in principle would prefer three dedicated sensors, it could make a lot of sense to replace half the green sensors in the Bayer pattern with unfiltered white.

[Graeme Nattress]

I don't see how this helps though, and the idea has been around for ages - I don't think it's new. I can see how it increases sensitivity, but that's not the problem with any sensor with decent sized pixels, it's dynamic range, and that "white" pixel will just saturate sooner. And yes, you could do an algorithm to guess what's going on in the unsaturated RGB situation, but we can do that already with the standard pattern.

I can also see it complicating colorimetry and making raw processing harder and bayer demosaicing a lot harder.

Graeme

[David Mullen ASC]

You're probably right and this new design approach will only be for some consumer camera products where speed is valued over color, hence why they are thinking of putting it into cell phone cameras and whatnot.

[Graeme Nattress]

I can see it being for such consumer products, yes.

[Graeme Nattress]

More info here:

http://www.dpreview.com/news/0706/07061401kodakhighsens.asp

So it's not just replacing a green with a white - it's a choice of 3 totally new patterns.

Graeme

[Bruce Allen]

I don't see how this helps though, and the idea has been around for ages - I don't think it's new. I can see how it increases sensitivity, but that's not the problem with any sensor with decent sized pixels, it's dynamic range, and that "white" pixel will just saturate sooner. And yes, you could do an algorithm to guess what's going on in the unsaturated RGB situation, but we can do that already with the standard pattern.

It increases dynamic range because the clear sensors will pick details out of the shadows when the sensors with color filters on top of them cannot, and the color-filtered sensors will give you highlight detail after the clear sensor has clipped.

I agree that it's not for pros. But one day soon you will have a system similar to Fuji SR where you have a mix of sensors sites of different sensitivities (due to their size, or ND filters, or whatever...). Either that, or current incremental improvements in signal-to-noise ratio / dynamic range will continue to the point where your dynamic range is "good enough". At the moment, it's the main disadvantage of shooting digitally.

Also, Graeme, I have been shooting some Stouffer tests of my own and am interested - how did you decide on the precise threshold between "I can see a 1/3rd stop of detail here" and "there's too much noise for this to really count"?

Bruce Allen
www.boacinema.com

[Bruce Allen]

More info here:

http://www.dpreview.com/news/0706/07061401kodakhighsens.asp

So it's not just replacing a green with a white - it's a choice of 3 totally new patterns.

Graeme

Thanks for the link Graeme! Looks great. It seem that they are doing heavy noise reduction on the color channels, maybe an intelligent sort of reduction driven by the white-pixel derived luma values?

A bit similar to what some of you did with your plugins years ago, no? G Nicer would fix the color values, partly informed by the info in the luma, right?

Bruce Allen
www.boacinema.com

[Guy]

What if you replaced the bayer color filter pattern with different levels of ND. Could you conceivably design an algorithm that would give you super HDR black and white?

[Graeme Nattress]

There's all sorts of things you can do..... But sometimes you want to do as little as possible.

Graeme

[Bruce Allen]

What if you replaced the bayer color filter pattern with different levels of ND. Could you conceivably design an algorithm that would give you super HDR black and white?

Yes, at the expense of ISO equivalent... Graeme's words are wise on this matter!

Bruce Allen
www.boacinema.com

[Axel Mertes]

Yes, at the expense of ISO equivalent... Graeme's words are wise on this matter!

Bruce Allen
www.boacinema.com

Its potentially not done by RED but things like this have been done in industrial use. I've seen 20 f-stops CMOS doing about this for night view car driving systems.

I wonder how Kodak gets a patent on something that has been done before...

Anyhow, there is an alternative technique called "dual slope" or "multiple slope" recording that is exposing once long, once short and adding the two frames into an HDR image. The lack is its two consecutive frames, but shot very fast after each other to minimize the issue. In such case the "ND" filter type might be more interesting, as it shoots at one time.

You can also create a new multiple RGB pattern with RGB pixels in different ND levels. That would be about the same what Kodak does, but requires a different decoding. I've seen a sensor a while back that had two levels of green exposure, so not BGRG but BG'RG", enhancing dynamic range already. Its not really new...

Axel

[Gavin Greenwalt]

Well someday if we had a 16k sensor we could also have .5ND bayer patterns spiced in. Once we have 32k sensors we can start to add microlenses and variable DOF... etc etc etc.

As resolution increases a lot of other physics problems can be solved. Somehow I doubt we'll continue to desire more than 8k of resolution unless it's going into an imax sized application. We're already on the bleeding edge of our eye's accuity with 8k.