Foveon sensor for digital cinema?

[Rudi Herbert]

Hello guys,

Just wondering if any of the technically savvy people on the forum know if the Foveon sensor has not, or could not, be considered for digital cinema. For those who don't know about it, it is a CMOS sensor without a Bayer or an anti-aliasing filter, resulting in images without virtually any noise or distortion. Also, it stacks 3 diodes, one atop the other, per pixel so that in effect it provides the advantages of the 3 sensor design on a single sensor. Each of the diodes reads a separate R, G or B value, so that each pixel effectively registers full RGB information.

I knew the sensor had seen limited use on some digital cameras, mainly SIGMA, but it wasn't until this weekend that I saw some images taken with it that I realized how beautiful the images truly are, and they do indeed look different from other sensors. We shot some comparisons with a Nikon d-200 and the SIGMA SD-14, with a 4.7 mp sensor, and to me, the SIGMA's were definitely richer and more naturally saturated in color. they had more of film's "organic" look that the Nikon, which to me is the best digital SLR out there. Now, since this sensor does not need a Bayer or any other filtering, couldn't this result in faster transfer rates, i.e. higher frame rates, that other CMOS cannot offer yet. I don't know, I found the images amazing, and the sensor a very intriguing technology.

Any comments?


Rudi Herbert

[Kevin Halverson]

The absence of an anti-aliasing filter (OLPF) is a very questionable practice. Any sampled system (not just digital imaging) should always have a LP filter to prevent aliases as once quantized, they can never be extracted from the resultant sample.

The Foveon is an interesting approach, but at this point in time, lacks the speed necessary to support cinema frame rates.

[Jeff Kilgroe]

This topic pops up here every couple weeks and there's already a few good threads on it. Try this one (http://www.reduser.net/forum/showthread.php?t=1358&highlight=Foveon).

[Graeme Nattress]

"without virtually any noise or distortion." - this is not the case. Indeed, the chroma on the Foveon is very noisy and requires noise reduction as part of the raw processing. The lack of OLPF causes obvious distortion, putting stair steps on diagonal edges.

"so that each pixel effectively registers full RGB information." - not exactly, unless you want an RGB image that looks like muddy brown. The layers are not RGB, but only labled with those colours - the actual spectral response is very different from RGB and intensive processing is needed to make nice colours, and this introduces more noise. Ouch.

Not having a bayer pattern is actually a disadvantage. Bayer is essentially a 1/3 compression scheme done on chip. So for the same output resolution Foveon files are a lot bigger.

Graeme

[Rudi Herbert]

Thanks all,

Apologies for starting a thread that had been discussed before at lenght. Thanks for the link to the previous discussion, however, I have a couple of questions for Graeme if I may imposs. We all know the Bayer is an old system, and to its credit, it has been refined to produce amazing results, but it does have its problems and they can be ameliorated but not solved. Now, the Foveon is probably the first step in trying to improve on the Bayer while still sticking to the reliable CMOS format. I like that. I respect that. Furthermore, the images I saw from the SD-14 were truly beautiful and organic. They were actually soft and pleasant, and they made the D-200 look unnaturally sharp for the first time ever. I did not see any of the jagged edges, even at 1600% magnification that the white papers would indicate should be expected...
Wouldn't an approach like the Foveon be improved into something that can produce similar, if not better, images than a Bayer? Jim said he would take a look at it when the Foveon can match the Mysterium, and he's right. But perhaps if they knew a company like RED is interested in their chip, could their engineers make the Foveon take the quantum leap and bring it to the level of Mysterium?
Then again, there's nothing wrong with the Mysterium, the Milk Girls and PJ's short look amazing, but I wonder if the future of the CMOS lies with a Foveon type approach or we'll simply accept that the Bayer is as good as we can do and leave it at that?

Thanks all for the replies,

Rudi Herbert

[Graeme Nattress]

Well, any sharply focussed Sigma Foveon image I've seen has been riddled with aliassing artifacts as they don't have an OLPF. I've never seen them look soft and organic because of all the aliassing artifacts which make the image look computer generated.

The other issue is the use of silicon as a colour filter, rather than the dyes that other sensors use. Silicon is a rather poor filter and this necessitates rather strong processing to get any kind of good colour out.

Graeme

[Graeme Nattress]

Here's a good example of the precise aliassing problems you get from avoiding the necessary OLPF:

http://forums.dpreview.com/forums/read.asp?forum=1027&message=23493121

Graeme

[David Mullen ASC]

The example is interesting -- you know, when I saw "Star Wars: Attack of the Clones" in IMAX, I saw similar artifacts, so I guess if you enlarge an F900 image that much, you may see something like that...

[Graeme Nattress]

When you enlarge any pixel based system, you're going to see aliassing artifacts. Unless you have a very smart interpolator.

Problem with the Sigma cameras is that they omit the OLPF so you get these problems without enlargement.....

Graeme

[Rudi Herbert]

Graeme,

Then, in your expert opinion, do you feel that at present time the Bayer is still the best solution to squeeze some magic out of a CMOS? I mean, obviously you guys have made an investment on the Mysterium with a Bayer on it, and judging by the results, it is doing just fine thank you very much. But let's not negate the fact that progress is inevitable, so if you were to be comfortably perched on your sofa a year from now, after RED has sold its first 10 000 units, what improvement/replacement for a Bayer would you dream of as you caress that cold drink in your hand?

with respect,

Rudi Herbert

[Graeme Nattress]

Bayer is currently the best compromise for our use. We have 3 choices:

3 chips with prism - can't use this as our chips are too big for this approach and we'd not be able to use standard cine lenses

1 chip with bayer - this is what we've chosen

1 chip with layers like Foveon. Currently doesn't have the frame rate or low enough noise.

I like the idea of the Foveon layered approach, but it looses out in terms of resolution to the bayer approach with how it's pixels are spread out, giving more spatial resolution in luma by sacrificing some chroma. So, if I'd change some things I'd make the lower layers on the Foveon style offset from the top layer, I'd have to find a better colour filter than just silicon and I'd definately use an OLPF.

Graeme