Red & Foveon?

[Chris Nuzzaco]

This is a curious question I've had bumping around in my head. Will later models of Red use a Foveon style CMOS sensor? So far, unless I've read wrong or been misinformed, Red is using a CMOS with a filter pattern (like all major DSLR's, with a few exceptions like the new Sigma SD14).

The Foveon CMOS sensors are three layer sensors (RBG). Thus they capture color more like film. People claim the colors are more true to life and the quality is supposed to be very high. Dynamic range specs, well, can't find any, but its touting to be as good as film.

Just a curious question for the Red Team.

:)

[Poi Boy]

I've been watching Foveon for the last six or seven years and they have always been promising in theory but thus far have not delivered. I'll keep watching.
Aloha
-A

[jeffry alan]

This question was brought up (by myself and others) a long time ago and (I beleive it was) Graeme went into some of the probelms with the Foveon technology.

Will post a link when I find it.

J*

[Chris Nuzzaco]

http://www.sigma-sd14.com/

sample images to look at ;)

I've been looking into their SD14 as a new camera body, but it won't take the Mighty Nikon F-Mount....:angry03: Oh well, their mistake...My biggest interest was the 24 fps! Probably don't need to explain why :P

[Andrew M.]

Foveon has long way to go in terms of resolving power and it is not near the CCD in terms of signal to noise ratio and repeatability of the signal on each photo-sites at different light levels. It is deep structure so I see focus problems ahead at 4K and up.

[Anders Holck]

The newest Foveon chip is only 5 fps at 2652 x 1768, and 30 fps at 640x480.
The Mysterium is 60fps at 4520 x 2540.

No competition

[Chris Nuzzaco]

The Foveon stuff might just be better suited to still work then.

Humm...I always like to tinker. A camera shop in my area has one of the SD14's, might go shoot some tests.

I'm a fan of using my D70 to make very high quality film clips. Works great if you know how to do it right (ie, proper lens setup to avoid exposure flicker, post processing workflow). Very good at time lapse also.

[Graeme Nattress]

If Sigma would put an OLPF filter on their cameras, I might be interested, but I can't stand the jaggies. Noise is also an issue - look at the Foveon white paper which show how the "RGB" sensor is processed as luma / chroma with heavy noise reduction on the chroma. Basically they get a luma from adding up all the photons in all the layers. Hardly an RGB sensor any more, is it, if you have to do all that interpolation and NR to get a nice image.

[Jeremy Hughes]

I heard Sigma had a high definition Foveon camera planned a while back. But never heard anything since.

[jbeale]

The Foveon sensor was an immediate sensation when it was originally announced because it seemed (theoretically) to have such a great advantage over the conventional Bayer approach. In practice, the implementations shown so far have been disappointing, and have gained little market share. Every professional and serious amateur photographer I know is using either Canon or Nikon DSLRs, both of which use the time-proven Bayer sensors. (edit: of those using digital SLRs anyway- I still see some medium-format film out there also, along with digital backs which are also bayer-type)

[Mark B.]

Every professional and serious amateur photographer I know is using either Canon or Nikon DSLRs, both of which use the time-proven Bayer sensors.

Shooters who already had Canon and Nikon glass didn't want to reinvest in a new Sigma mounted camera... that was probably the main reason that Sigma sales didn't pick up. That and some quirky design flaws in the Sigma camera body. But the Foveon chip wasn't one of the aspects that detracted from the camera.

A lot of people, myself included, prefer the foveon image sharpness over soft-focus bayer images.

[Graeme Nattress]

Ah, but what you're seeing are aliasses and false detail, not sharpness. There's a difference. Sharpness is real and aliasses are artifacts that looks like sharpness but not.

On the other hand, bayer pattern images are anti-aliassed and don't have unnatural jaggies or aliasses in them, because they use an optical low pass filter. Sampling theory tells us that if you sample an image / audio / whatever, and don't low pass filter, you'll get artifacts.

Now, to an extent, you can "get away" with that in a still image, as if any aliasses are objectionable, you can go paint them out. But with moving images, aliasses look twice as ugly, and you can't go paint them out every frame....

Now, Foveon is up to 4.7mp, whereas bayer patterns sensors are 10,12,16, or even 39mp. With 10mp and above, you can downsample the image to the size the Sigma makes and get it looking just as sharp and probably with less aliassing too.

Now, if Sigma installed on OLPF, I might be interested, but I cannot stand jaggies in my images.

It's also funny that Foveon claim that to make a bayer image into RGB you have to interpolate. This is true, but to turn a RAW Foveon image into RGB, you have to, guess what, interpolate - not between pixels, but withing the R, G and B of that pixel, and to a much greater extent than you have to with bayer image. A RAW Foveon image looks almost black and white, whereas a RAW bayer image just looks a little under-saturated. Because of the amount the colours in a RAW Foveon image have to be boosted, you get noise, and hence (if you take a look at the Foveon white papers) noise reduction is a key part of the RAW conversion algorithm.

Both Foveon and Bayer CFA are compromises, just as the 3 chip prism system is a compromise. There is no perfect image sensor and no perfect compromise.

But the good thing is, is choice, and you can pick the compromise that best suits your visual system. Because if you like the image, it doesn't matter how it was made.

Graeme

[Bruce Allen]

Graeme, I'm sure you know these, but...

1. Foveon doesn't need an OLPF as much because its moire is only in the luma. With Bayer, you get nasty color moire artifacts too, so there is more of a need for an OLPF.

2. Some Bayer-sensor cameras such as the Kodak 14n do not have an OLPF.

3. In the future (eg 30 mp 35mm sensors) you might not need an OLPF because you will possibly be lens-limited (the lens will act as the OLPF).

4. To characterize OLPFs as either "off" or "on" is wrong - you can have an OLPF that is too weak (you still get some moire), or too blurry. I think almost all OLPF designs trade off a slight chance of moire in exchange for some sharpness, then use the RAW conversion algorithm to remove most residual moire.

5. The Foveon is not interpolating spatially.

6. Foveon has problems anyway though because the sensor at the bottom of their RGB stack gets too little light and is hence noisy.

I totally agree about sensors being a compromise though and think that Red is making the best compromise by far.

Cheers

Bruce Allen
www.boacinema.com

[jbeale]

> 3. In the future (eg 30 mp 35mm sensors) you might not need an OLPF because you will possibly be lens-limited (the lens will act as the OLPF).

I'd say, on any system with an interchangeable lens system you are likely to use lenses with differing degrees of resolving power at different times, so I'm not sure if relying on the lens as an optical LPF is a good system design. Also, there is always a tradeoff involving sensor dynamic range with pixel size and other things being equal, you get more dynamic range with larger pixels, so I don't see a reason to make your pixels smaller than necessary.

I'm sure there are other factors at work besides just pixel size, but looking at the relative noise levels in sample full-scale image frames from the Viper vs. Red, there is quite a contrast.

[Graeme Nattress]

Bruce,

1) um, yes you do. It's basic sampling theorem. With Bayer it's especially important as to avoid chroma moire, but without an OLPF, on a Foveon you still get luma moire and aliassing, which is very visible. Read some of the Foveon white papers and they will say, hidden in the text that you still need low pass filtering.

2) Indeed, and I'd say that makes them also incorrectly engineered.

3) Yes, as the pixels get small enough that the lens acts as an OLPF.

4) Yes, there are varying degrees of filtering. Doesn't mean that you can "get away" with none without visible artifacting from it, even on Foveon.

5) Correct, it's interpolating depth wise so to speak.

6) Yup. They basically sum all layers to get a luma that's relatively noise free, then do some noise reduction magic on the chroma. Re 5) above, the noise reduction may have a spatial component - the white paper doesn't go into details of the NR needed, but NR tends to be either spatial or temporal. Given that you're taking one image that sort of rules out temporal. (Yes, could be temporal taking a black frame, but that's only usual on "long" shots).

Yup, the choice of compromise can be a tricky one. I think we've got the balance right, but it's good to understand the alternatives too.

Graeme

[Jannard]

When Foveon can match the resolution, frame rate, lack of noise and "feel" of the Mysterium... we'll take a look.

Jim

[Mark B.]

Ah, but what you're seeing are aliasses and false detail, not sharpness. There's a difference. Sharpness is real and aliasses are artifacts that looks like sharpness but not.


My thoughts are that the Bayer's estimation of how much color MIGHT be at a neighboring site is as much or more of a false detail than the single-point collection method used by the Foveon. And sharpness is really a property of the thing being looked at more than the item it represents... or, at least that's how I was using the term. If a picture has high contrast between pixels, even if it's from aliasing effects, I still consider it to be a sharp picture. Maybe I need a different term than sharp... crisp maybe, or contrastic. There's a new word... 'contrastic'. Or we could try 'focustic'.

"I was taking a picture down by the docks and I got a really focustic picture. Produced a very contrastic print. Course then I cut my finger on the edge of the photo, because it was really sharp."

Ah well, we've already done this whole Foveon/Bayer discussion before and it would be a waste to do it all over again. If Foveon doesn't work for the video camera, then it doesn't work for the video camera. The Bayer moire makes me nervous but hopefully it won't be evident with the image changing all the time.

[Graeme Nattress]

It's not evident at all - it just looks like a good image. As I said, the perceived sharpness of the foveon is as much a factor of the lack of necessary OLPF filter as anything else. You just can't get away without one on a moving image as you'd get twitter, jaggies, moire and all other nasties you just don't get with film, or well done video for that matter.

Graeme

[Chris Nuzzaco]

Wow, good stuff guys. Did the idea for RED stem from using DSLR's for stop motion and time lapse? When I get some time, I'll upload a link of some stuff I've done with a D70. I have to say, its pretty breathtaking what you can do with one if you get the approach nailed down right. The first thing I noticed, almost zero grain...I had to add grain to make it look more like film, it didn't even look like HD, too clean for that. Of course, a lot that has to do with exposure, lighting, camera settings, and post processing too, but mine have always looked super clean.

[Bruce Allen]

Thanks Graeme and Jim for your excellent replies to my not-very-politely-worded post. Happy weekend!

Bruce Allen
www.boacinema.com

[Radoslav Karapetkov]

Doesn't an image, captured with a Foveon type sensor, require a Foveon-like kind of display so that these (supposed) advantages can become really visible? :detective2: