Foveon Chip for Scarlet

[Norman B. Miller]

Just viewed some images for the new Foveon X3 chip (Although they are probably at 5 or 6 now) I was very impressed with the depth of color and color rendition, and resolution....Was wondering what the pros here thought about the use of the Foveon chip for Scarlet....Thanks...

 

[Stephen Gentle]

Just viewed some images for the new Foveon X3 chip (Although they are probably at 5 or 6 now) I was very impressed with the depth of color and color rendition, and resolution....Was wondering what the pros here thought about the use of the Foveon chip for Scarlet....Thanks...

I'm sure they have thought about it... As for using it in Scarlet, at this point in the game, it's impossible unless you want it to take another few years, and cost a lot more because of all the R&D of making a new sensor... There are probably other upsides to using the Bayer pattern or they wouldn't have chosen it anyway...

 

[Robert Jackson]

I'm sure they have thought about it... As for using it in Scarlet, at this point in the game, it's impossible unless you want it to take another few years, and cost a lot more because of all the R&D of making a new sensor... There are probably other upsides to using the Bayer pattern or they wouldn't have chosen it anyway...

I'll bet 2/3" Foveon would have been very possible. It probably wouldn't have been as sensitive in low light, but at ISO 100 or so I'll bet it would have looked killer. And 3K would have been 3K.

 

[Graeme Nattress]

That's only because they omit the OLPF and allow though some pretty viscous aliasing artifacts. Also, because the colours in the raw are very poorly differentiated, you need very aggressive matrixing to bring them up to saturation which has a rather negative impact on the noise floor and colorimetry. There's no such thing as a free lunch...

Graeme

 

[Daniel Browning]

This has come up before:

Foveon Sensor for RED?

It goes into more detail about the reasons why Foveon is inferior to Bayer:

• Noisy chroma
• Aliasing artifacts
• Poor color accuracy
• Low resolution

Just viewed some images for the new Foveon X3 chip (Although they are probably at 5 or 6 now)

X3 refers to the number of color samples per pixel, so they're still on X3. Here's the history:

• 2002: Sigma SD9, 3.4 MP
• 2003: Sigma SD10, 3.4 MP
• 2007: Sigma SD14, 4.7 MP

While it may seem like they are dragging their feet, that's not it. They're going very fast for someone who is swimming against the current. The reason Bayer sensors have advanced much much faster is because 99% of the technology investment is there. (And the reason for that is because Bayer is objectively superior.)

I was very impressed with the depth of color and color rendition, and resolution

You were misled. The "resolution" is in fact just a side effect of removing the anti-alias filter. Foveon can only sample luma and chroma at to 51 lp/mm (using 2.5 samples per line pair). A six-year-old DSLR (300D) can beat that for luma (54 lp/mm) and a modern DSLR (500D) blows it away (85 lp/mm) for luma can nearly match it for chroma (45 lp/mm). Of course, additional chroma resolution is invisible in most everything except test charts, and whatever difference there is disappears as soon as you export a JPEG.

The color rendition of Foveon is a combination of very poor color accuracy, lots of processing, and aliasing.

If you want a "Foveon" image, just create one in post from a Bayer image:

• Match the high Foveon chroma noise by adding a bunch of chroma noise in post.
• Match the poor Foveon color accuracy by adding junk to the colors in post.
• Match the low Foveon resolution by downsampling the Bayer image from 14.7MP (or whatever) to 4.7 MP.
• Match the Foveon aliasing artifacts by looking for a downsampling program that does not have an anti-alias filter.

In fact, someone actually did do just that:

Software for simulating Foveon with a Bayer image

Of course, going from Foveon to Bayer is impossible.

Foveon was a nice idea that didn't work out.

 

[Lawrence Bansbach]

Just viewed some images for the new Foveon X3 chip (Although they are probably at 5 or 6 now) I was very impressed with the depth of color and color rendition, and resolution....Was wondering what the pros here thought about the use of the Foveon chip for Scarlet....Thanks...

Foveon is a nonstarter, totally unsuitable for HD video, let alone 5K or even 3K. It was an interesting technology five years ago, but it never really went anywhere.

 

[Wesley Scoggins]

Foveon sensors while interesting, I personally feel are useless without Foveon based displays with stacked pixels, so they can naively present the footage without having to separate the colors for the individual pixels.

 

[Frank Jonen]

@Lawrence Foveon chips are being used in security and industrial cams with HD res. The only reason why you haven't seen moving footage from it is because these are impossible to use on standard tripods or shoulder mounts. Basically when you can't afford a gross misreading of colors due to debayering. Image clarity and res is not such a huge factor there.

One thing I liked when shooting with a Leica M8 was the clarity you get when you don't anti-alias, it's like removing a protective foil off an image of compared to even the most recent Nikon and Canon cams. Also makes the noise harder to deal with as it's now made of single pixels instead of some smoothed blurbicle.

Would be great to have that option in the Scarlet. Say: "don't anti-alias this I have other plans." or "Yes anti-alias this, I don't care".

 

[Axel Mertes]

If it was possible to capture RGB in a single pixel, then nature would have invented it.

In fact, the human eye uses a pattern too, not particularly a Bayer pattern, but a pattern, consisting of a mix of red, green, blue and just luminance sensitive "pixel elements".

The main reasons why a human eye has few aliasing issues is that

a) the brain has probably some wonderful unknown demosaic algorythms we wish we would understand or know of, something like a "neuronal net approach"
b) their is no fixed structure in the pattern, making the pattern itself "noisy" and thereby almost free of moire effects common in sensors - this pattern leads to some pretty well working sharpening "between the lines" as there are no lines at all... if you understand what I mean. The human eye can reveal resolutions beyond its physical resolution, due to "sub pixel" information handling (time samples, motion jitter).

There was (or maybe still is) a project by IBM or Intel called "Super Resolution" where they analyzed shaky old video/film frames to extract superior resolution textures and 3D models of the scene to finally render out a HD or higher resolution video out of PAL/NTSC or S8 material for instance. Amazing! Works only with "shaky" information, typically derived from small jitter of hand camera or film cameras Malteser cross.

IMHO a pattern like the Bayer pattern is a very valid and clever approach on capturing color information for an image. As clever as Foveon might look like in the first place, it has a lot of weak points, mainly sensitivity.

What we need it invent is a chip that not just counts photons, but although their wavelength (without using filters). No such thing on the horizon yet AFAIK...

I see no reason why RED should start using differen sensor designs. Maybe they should drop on rolling shutter at some point (this is a really weak point), but all the rest can be accomplished with increased resolution and sensitivity.

Cheers,
Axel

 

[Norman B. Miller]

Thank You

Thank You

Thanks to all of you for your astute observations on the Foveon sensor....The level of talent on this forum is quite amazing. I'm very much looking forward to getting my Scarlet.

 

[Chad Clemans]

Was wondering what the pros here thought about the use of the Foveon chip for Scarlet....Thanks...

From what I've read about Foveon sensors, the luminosity levels that permeate down to the lowest layer (red) are depreciated by passing through the upper levels; that's its issue.

With Bayer sensors the luminosity seems to be gathered more accurately but the selection of which color a pixel is supposed to be, around the edges of objects where the color changes, is not as accurate; that's its issue.

Also from what I've read, and there are discrepancies, the human eye can see up to 16k.

Therefore as RED is ever increasing resolution, and therefore cramming more pixels in there, thus making the guessing of colors around the edges of objects more and more negligible to the human eye, and because Bayer sensors seem to be really good at halving resolutions when needed, I'd go with Bayer for now. I haven't heard of many solutions for the Foveon luminosity problem.

 

[Dan Hudgins]

OLPF

OLPF

I don't see a reason why you can't use an IR/OLPF filter on the Foveon sensor just as you would on a monochrome CMOS sensor to reduce aliasing.

I like the Sigma Foveon images better than any Bayer images I have seen, they look more like film. And the absence of chroma moire. Even with the heavy OLPF the RED ONE images show chroma moire when you increase the chroma matrix on some underexposed shots, because of the REDCODE they get blured into patches of off color on textured subjects this limits how much you can bring the color up in dim dark shots.

Maybe the dynamic range is an issue though.

A bigger issue is the interlayer Capacitance, I don't think you will see any Foveon of 2K running faster than 60fps or even 24fps soon or ever. The Foveon are too slow to be used for a Digital Cinema Camera, and even though a 2K bayer filter sensor does not resolve 2K like a three chip camera could, a 3x1K Foveon would not have enough luma detail for 2K filmout, especially if you use a heavy enough OLPF to blur the image with those large pixels to reduce the aliasing.

Another issue for a true RAW camera is that being true RGB (well white, yellow, and redish, but you end up with RGB and still need to save all three signals) you end up with 3x the data to record, so you need 3x the bandwidth and 3x the recording media.

 

[Miguel "Macgregor" De Olaso]

This has come up before:

Foveon Sensor for RED?

It goes into more detail about the reasons why Foveon is inferior to Bayer:

• Noisy chroma
• Aliasing artifacts
• Poor color accuracy
• Low resolution

X3 refers to the number of color samples per pixel, so they're still on X3. Here's the history:

• 2002: Sigma SD9, 3.4 MP
• 2003: Sigma SD10, 3.4 MP
• 2007: Sigma SD14, 4.7 MP

While it may seem like they are dragging their feet, that's not it. They're going very fast for someone who is swimming against the current. The reason Bayer sensors have advanced much much faster is because 99% of the technology investment is there. (And the reason for that is because Bayer is objectively superior.)



You were misled. The "resolution" is in fact just a side effect of removing the anti-alias filter. Foveon can only sample luma and chroma at to 51 lp/mm (using 2.5 samples per line pair). A six-year-old DSLR (300D) can beat that for luma (54 lp/mm) and a modern DSLR (500D) blows it away (85 lp/mm) for luma can nearly match it for chroma (45 lp/mm). Of course, additional chroma resolution is invisible in most everything except test charts, and whatever difference there is disappears as soon as you export a JPEG.

The color rendition of Foveon is a combination of very poor color accuracy, lots of processing, and aliasing.

If you want a "Foveon" image, just create one in post from a Bayer image:

• Match the high Foveon chroma noise by adding a bunch of chroma noise in post.
• Match the poor Foveon color accuracy by adding junk to the colors in post.
• Match the low Foveon resolution by downsampling the Bayer image from 14.7MP (or whatever) to 4.7 MP.
• Match the Foveon aliasing artifacts by looking for a downsampling program that does not have an anti-alias filter.

In fact, someone actually did do just that:

Software for simulating Foveon with a Bayer image

Of course, going from Foveon to Bayer is impossible.

Foveon was a nice idea that didn't work out.

So what if any, are the advantages of a foveon sensor?

 

[Dan Hudgins]

Chroma moire

Chroma moire

So what if any, are the advantages of a foveon sensor?

Mostly its the lack of chroma moire, it lets you get chroma and luma bandwidth that are equal, in a Bayer camera the chroma is always lower than the luma to suppress the chroma moire.

A lack of chroma moire is more important in a still camera, in a video camera the compressed distribution has lower bandwidth in the chroma so you would lose the advantage.

 

[Daniel Browning]

I don't see a reason why you can't use an IR/OLPF filter on the Foveon sensor just as you would on a monochrome CMOS sensor to reduce aliasing.

Agreed. I think Sigma chose to go against the recommendations of the Foveon designers and remove the OLPF because it saves money and lets them tap into the market of people who love luma aliasing artifacts.

Even with the heavy OLPF the RED ONE images show chroma moire...

Yeah, chroma aliasing can sometimes be a problem with Bayer, and demosaic algorithms can be adapted to help deal with it (e.g. RED's "noise reduction" anti-more modification). I think it's a pretty minor issue.

So what if any, are the advantages of a foveon sensor?

In 2002 it could compete a little better, but billions of dollars R&D on Bayer CMOS have left Foveon CMOS in the dust (resolution, low light, etc.).