FOVEON - Dead in the Water?

[Joseph S Wisniewski]

Bayers have had a few generations to improve,
Foveon (using the term loosely) is really in its infancy.

I thought we covered this already. Layered sensors didn't begin with Foveon, they actually predate Bryce Bayer's work. Many of the companies with Bayer (or other filter array pattern) sensors in production today also have research (published and unpublished) and even patents in layered sensors.

Even
Sony are trying to find other ways to improve colour in CCDs with
their Cyan detector.

A very interesting development indeed, since 3 color (i.e. Bayer) filter sensors already outperform silicon depth color separation sensors (i.e. Foveon) in terms of color accuracy.

Given it 5-10 years, we'll all be using, or wanting 'Foveon' type
design (even if its not Foveon assuming they don't survive...) in
our Pro cameras, and be ashamed at the quality we used to think was
great!

I think it's a basic SITS problem, and the known drawbacks of filter array sensors are best solved by multiplicity. Make the CFA cells smaller, and you can solve the chroma aliasing problems by using cells small enough so that the lens resolution limit becomes the antialiasing filter (in which case CFA and Foveon sensors are on equal footing) or by employing alias-proof designs such as pseudorandom filter layouts.

The solutions to the Foveon noise and color accuracy problems are not so simple.

--
Ciao!

Joe

http://www.swissarmyfork.com

 

[Stan]

...to help give Bayer images more of the "3D Look" of a Sigma image. He says that Bayer cameras have inconsistent "focus" in the plane of focus since some items were sensed with fewer pixels than others as a function of color. Apparently this came about after his efforts comparing the Sigma SD10 and a Canon 10D. Worth a look, and there are sample images on his site and in the Sigma forum (search).

From the Qimage help file, see 2nd paragraph:

"About the Qimage unsharp mask: The radius parameter equates to the width of the sharpened edges. The larger the radius, the wider the edges will appear. The strength parameter is the contrast of the edges so that the higher the strength, the more contrast the edges will have. Typical values for radius and strength start at about 1/100 or 2/100 respectively so these are good starting points. Note that Qimage does not use a "threshold" parameter since sharpening is performed on the luminance channel only, avoiding noise and color fringing that the "threshold" is meant to address.

Sharpening equalizer: Due to the way most (Bayer based) image sensors capture images, some colors in the image may appear sharper than other colors. For example, black, white, and gray details may appear sharper than red, blue, or green details because fewer pixels on the sensor are contributing information for those colors. Moving the sharpening equalizer to the right will sharpen colors more than black, white, and gray areas, thereby compensating for the sharpening discrepancy of images captured by most digital cameras. With the equalizer at max, use a higher level of sharpening such as 2/200 to see the effect."

Stan

 

[Sean Keegan]

Why not call it what it is. The very best 3mp camera ever made, with qualities that make it "comparable" to 6mp DSLRs. Of course both systems have their flaws and strengths, but if Foveon would get their act together and make sensors at the same MP ouput size as Bayer there would not be any of these arguments. A 6mp X3 would blow away a 6mp bayer. But right now, there is no reason to get excited over X3 technology. You did the right thing buying into the Nikon lens system over the Sigma lens system. Would it be nice is there was a Foveon based Nikon DSLR? Sure, but there is no pressure to do so when the stuff Nikon currently has is just as good. That is why Sigma is doomed to be the only X3 player for some time. Fuji has their own amazing Super CCD, Nikon now has LBCAST, and Canon has their wonderful CMOS sensors. They are all supurb. What does Foveon bring to the table. Nothing really at the current state of things.

Regards,
Sean

So it would be even only 1/4th of your 6 Megapixels single 3-color
pixels as far as I see

This would mean that Nikons new D2H and the Canon 1D are in fact 1
MP cams LOL .

Anyway... I think we all just count the OUTPUT megapixels... there
is more than only one way to get there (interpolation, foveon, etc).

Cheers

Just remember your supposed image created from your 6MP sensor
(jpg/tiff etc) has 6MP pixels with Red, Green and Blue data, yet
you only captured one of those. You made the rest up by guessing...

My long term prediction is that once these basic sensor designs are
able to meet the noise expectaions of most at higher ISOs, the
technology will progress to the same direction that Foveon have
already gone.

It happened with video cameras, it will happen with still cameras,
and its likely to be the biggest difference in the future that
seperates Pro cameras from consumer cameras.

Peter,

Foveon still hypes their sensor on their website as being 3x the
actual Megapixel rating, which is shear fantasy and delusional
thinking. It was part of this kind of thinking, that got Foveon
where they are today. I was once a proponent of the SD9, but after
seeing it in use, I walked away unimpressed and with a little
resentment on the marketing aspect. The SD10 is only prolonging the
death of Foveon.

If Foveon can make a 5fps, ISO 200-3200 10MP (actual pixels, not
ones that exist in FoveonLand) chip, and put it in a Full-Frame
DSLR, they will come back with a punch.

But unfortunately, these companies don't listen to real users, and
instead think inside a vacuum.

--
http://www.digitaldingus.com
http://www.digitaldingus.com/forums

 

[Joseph S Wisniewski]

Why not call it what it is. The very best 3mp camera ever made,
with qualities that make it "comparable" to 6mp DSLRs.

My personal belief is that Sigma isn't good enough with words or pictures to make that work in their adds. They seem rather reluctant to hire good European or American advertising people, so they end up with barely comprehensible ad campaigns.

Of course
both systems have their flaws and strengths,

OMG, someone understands

but if Foveon would
get their act together and make sensors at the same MP ouput size
as Bayer there would not be any of these arguments. A 6mp X3 would
blow away a 6mp bayer.

Yes, it would. But there's three things you have to keep in mind.

First, care and feeding: the 6MP Foveon would generate 18 "megasomethigns" of output, which means 13 megabyte (after compression) raw files. That's a lot of data to shuffle around, and the CF cards are going to fill up fast.

Second, that's a lot of data to process, without a redesign of the camera's electronics. Right now, the SD9/10 processor is just fast enough to get 1.9 frames/sec out of 3.5mp. Iif you bump mp to 6, you drop fps to 1.1. Increase the speed of the electronics to compensate, and you increase cost.

Third, it means an increase in sensor size, from 1.7x to 1.3x. Now, a lot of photographers would love this. But you just have to remember that if you nearly double sensor size, sensor cost is going to increase. It's probably going to more than double (weird relationship between chip size and "yield", the number of ships that turn out good).

So there's really nothing stopping this from happening, except cost. The SA-9 body used in SD9/10 would be a bit iffy for a "premium" model, but Kodak manages to muddle along with the Nikon N80 for their 14n, so it is something the market will bear.

Of course, my sources say we're going to see an under $2000 APS sized 10mp Bayer from one of the "big 4" come PMA. So that would leave Foveon still fighting the 6mp vs 10mp issue.

But right now, there is no reason to get
excited over X3 technology. You did the right thing buying into the
Nikon lens system over the Sigma lens system. Would it be nice is
there was a Foveon based Nikon DSLR? Sure, but there is no pressure
to do so when the stuff Nikon currently has is just as good. That
is why Sigma is doomed to be the only X3 player for some time. Fuji
has their own amazing Super CCD, Nikon now has LBCAST, and Canon
has their wonderful CMOS sensors. They are all supurb. What does
Foveon bring to the table. Nothing really at the current state of
things.

Not "nothing", but maybe "not enough". If Foveon can get into a large Asian fab, it should be able to beat the competition in price. That either means bigger sensors for the same money, or the same size sensors, cheaper.

Not going to tempt Nikon, Canon, or Fuji to give up their homegrown technology, but it sure might tempt Pentax, Minolta, or Oly into buying Foveon sensors.

But I don't think, with Nat Semi as the only fab, they will get any big design wins.

--
Ciao!

Joe

http://www.swissarmyfork.com

 

[Sean Keegan]

Hey Joseph,

but if Foveon would
get their act together and make sensors at the same MP ouput size
as Bayer there would not be any of these arguments. A 6mp X3 would
blow away a 6mp bayer.

Yes, it would. But there's three things you have to keep in mind.

First, care and feeding: the 6MP Foveon would generate 18
"megasomethigns" of output, which means 13 megabyte (after
compression) raw files. That's a lot of data to shuffle around, and
the CF cards are going to fill up fast.

I own a S2, so I am already dealing with 11MB files with no problem. It is not that bad, and huge CF cards are already out.

Second, that's a lot of data to process, without a redesign of the
camera's electronics. Right now, the SD9/10 processor is just fast
enough to get 1.9 frames/sec out of 3.5mp. Iif you bump mp to 6,
you drop fps to 1.1. Increase the speed of the electronics to
compensate, and you increase cost.

Again the S2 handles 11MB RAW at 2FPS with 8 image buffer, and I am sure with the improving electronics there could be improvment on that.

Third, it means an increase in sensor size, from 1.7x to 1.3x. Now,
a lot of photographers would love this. But you just have to
remember that if you nearly double sensor size, sensor cost is
going to increase. It's probably going to more than double (weird
relationship between chip size and "yield", the number of ships
that turn out good).

Very true, not to mention that we would lose our free longer telephotos

So there's really nothing stopping this from happening, except
cost. The SA-9 body used in SD9/10 would be a bit iffy for a
"premium" model, but Kodak manages to muddle along with the Nikon
N80 for their 14n, so it is something the market will bear.

Then why have they not done it? Perhaps it is harder to scale up their chip than we realize. Who knows. They should have raised the resolution a bit in the SD10, maybe to 4 or 5 mp in my opinion. Staying at 3mp was a mistake, unless they have something waiting in the wings.

Of course, my sources say we're going to see an under $2000 APS
sized 10mp Bayer from one of the "big 4" come PMA. So that would
leave Foveon still fighting the 6mp vs 10mp issue.

My point exactly. If Foveon really wants to be a revolution than they must beat bayer so badly no one can even compare them. MP for MP the X3 has a chance. Will Foveon ever get there? Does not look likely.

But right now, there is no reason to get
excited over X3 technology. You did the right thing buying into the
Nikon lens system over the Sigma lens system. Would it be nice is
there was a Foveon based Nikon DSLR? Sure, but there is no pressure
to do so when the stuff Nikon currently has is just as good. That
is why Sigma is doomed to be the only X3 player for some time. Fuji
has their own amazing Super CCD, Nikon now has LBCAST, and Canon
has their wonderful CMOS sensors. They are all supurb. What does
Foveon bring to the table. Nothing really at the current state of
things.

Not "nothing", but maybe "not enough". If Foveon can get into a
large Asian fab, it should be able to beat the competition in
price. That either means bigger sensors for the same money, or the
same size sensors, cheaper.

Not going to tempt Nikon, Canon, or Fuji to give up their homegrown
technology, but it sure might tempt Pentax, Minolta, or Oly into
buying Foveon sensors.

But I don't think, with Nat Semi as the only fab, they will get any
big design wins.

Yep, sad in a way, the technology has such potential. But so far I find myself agreeing with the title of this thread. Foveon is dead in the water. It is great technology, but not keeping pace with the rapid advances in the Bayer world right now. I kept expecting Foveon to get into the P&S market where they could make a lot more money, and build some capitol to push their technology forward. But it looks like Foveon is stuck in a very niche market with the Sigmas, and losing valuable market share every day.

Regards,
Sean

 

[Stan]

Why not call it what it is. The very best 3mp camera ever made,

with qualities that make it "comparable" to 6mp DSLRs.

Somehow it seems to be better than that at times. There's a thread in the Sigma forum right now where a guy made some A0 prints, poster size.

He upsized the image to 14 MB via the double size option in the raw converter and then upsized it to 48 MB via bicubic and it came out great.

I have heard some say that the Sigma/Foveon image upsizes pretty well since it has already had so little interpolation or words to that effect. You can't do that as well with the common 6 MP units out there, maybe hte Fuji could.

Stan

 

[DMB]

Hi, all,

Over a year ago, I was extremely heartened by all the advantages
that the new Foveon technology was going to bring to digital
photography. It promised to rectify most of digital's negative
aspects, or at least ameliorate them. However, I have heard nothing
about this promising technology for a long time -- save for the
Sigma D camera connection. Are major camera companies like NIKON
looking at Foveon seriously or is Foveon effectively another hyped
product that is destined to die? Anybody have an ear to the ground?
Was the "hype" just that? I hope not...
--
Peter Rackham

http://www.pbase.com/rackham/myview01&page=all

Not saying the performance is better on one vs. the other, just thought that with all the debate of which technology is superior, that some might find this exercise insightful.

http://www.outbackphoto.com/artofraw/raw_05/essay.html

Enjoy!
-----------------------------
Dave Boxmeyer

 

[Sean Keegan]

Hey Stan,

I agree the Sigmas are amazing in poster size prints considering the ouput file of 3MP. Stunning in fact. But here are my experiences printing Sigma files and comparing them to my S2 prints:

1. Aliasing. This is already pretty visable even at 3mp size. When I size it up to 12mp it becomes greatly exaggerated. The S2 at 12mp has very little aliasing, so I can enlarge up even greater before aliasing starts to show up in the large size prints.

2. Noise. Again, visable even at 3MP. When resized up to 12mp the noise gets larger, even though it is the same amount of noise, the human eye can see it easier. The S2, even at 12MP, is one of the lowest noise cameras on the market at any price, especially at high ISO. So again, one can blow it up a lot larger before the noise becomes apparent.

3. Resolution. I have done many tests with SD9 and S2 and the S2 is definitely a lot better in many cases, and about the same in other cases, such as color changes between red and blue. But overall the detail I get is much greater the majority of the time.

Keep in mind that if I did not have the S2 to compare the prints to, I would be stunned at the quality of the SD9 prints. So I can see how someone would be thrilled with the output of the SD9. But next to the S2 images of the same size it is obvious which has the better image quality. It is all relative. Great huge prints are possible with the SD9, but even larger and better prints are possible with the S2.

The argument that there is less interpolation to start with so one can interpolate larger is not entirely true either. I mean, honestly, when you look at bayer images, do you often see artifacts that can be attributed to bayer interpolation gone wrong? It happens in very special cases, but it is rare in the real world. Usually the images look great, with no color moire, or weird resolution artifacts. In fact I have found that it is pretty hard to fool the camera into producing errors, unless you choose a blue/red resolution chart.... but how often does that kind of thing happen in the real world.

Regards,
Sean

 

[Stan]

What happens to the jaggies after going through SPPs doublesize algorithm...still there, more, less?

You know what? Lots of Canon folks think that the 10D is the DSLR noise champ. It's not. The Fuju S2 is the champ. And for some reason it seems like the 12 MP mode was glossed over in the DPreview. Like it's a nonissue. Yet that Russian site really put it through a throrough test and it did very well.

Maybe the Fuji and Sigma can't get no respect!

Stan

 

[Sean Keegan]

What happens to the jaggies after going through SPPs doublesize
algorithm...still there, more, less?

Good question. I would be curious about this.

You know what? Lots of Canon folks think that the 10D is the DSLR
noise champ. It's not. The Fuju S2 is the champ. And for some
reason it seems like the 12 MP mode was glossed over in the
DPreview. Like it's a nonissue. Yet that Russian site really put
it through a throrough test and it did very well.

Yeah, the Russian site is great! Just amazing amounts of detail gone over and really scientific. I also liked the same site's comparison of digital to film, again with the S2. I wish there were some reviews as detailed as that in english, but I can imagine the time involved was immense.

Maybe the Fuji and Sigma can't get no respect!

Yeah sometimes that is true. I would realy like a 6mp Foveon based camera! Or even 8MP. The Fuji has its share of problems too though, like color moire' showing up a bit more often than I would like. Luckily it is pretty easy to correct, but still a pain. And despite having pretty much the lowest noise of any DSLR out there, that is really only the case if you get the exposure right. If you have to pull up an underexposed image there can be some noise that starts to show up in the shadows sometimes. And I wish there was a 3MP Foveon body that used Nikon lenses. Then I would probably have both cameras right now.

Regards,
Sean

 

[Sean Keegan]

Hey Stan,

Just in case you were curious, I put together an animation of the resolution chart on the Russian site. I find it is always easier to compare different cameras by overlying them and switching between them. The differences are more obvious then if you just use your eye and memory. Here it is doubled in size using nearest neighbor to make it easier to see... interesting huh?

http://www.pbase.com/image/24255129/original

And here is a similar animation of SD9 VS S2 using Phil's charts on this site. I blew up the SD9 to the S2 size using bicubic, and then cropped and doubled using nearest neighbor to make easier to see. The S2 has higher sharpening applied for some reason, I could not find one with sharpening off, but it does not matter in this example.

http://www.pbase.com/image/24246215/original

and here is a demonstration how two images of different size with the same amount of noise will have drastically different amounts of noise when blown up to the same sizes that I was talking about. This demonstrates what I have observed with the S2 vs SD9 images I have printed. But even worse as the SD9 has worse noise than the S2 at its native image output size. I am sure you can see that the same would be true of aliasing.

http://www.pbase.com/image/24312816/original

Regards,
Sean

 

[Stan]

...poster size, first doubled in SPP then bicubic. Check out the guy putting his face up to the poster in amazement. (Sony camera for this image) I don't know if SPP helps out here or not..I mean the Foveon folks were able to set it up to suit their chip.



Some 100% crops...check out the detail..Sigma camera

http://forums.dpreview.com/forums/read.asp?forum=1027&message=7183204

More, check out the contact lens!!

http://forums.dpreview.com/forums/read.asp?forum=1027&message=7183312

Stan

 

[Stan]

..are supposedly available or soon to be. I think the Nikon adapter is available now and the Canon is coming soon.

But even worse as the SD9 has worse noise than the S2 at its native image output size. I am sure you can see that the same would be true of aliasing.

That's why I was wondering about what is up with the double mode that takes you from 3.4 MP to about 14 MP using Foveon devleoped algorithms.

I've seen some other interesting images using that mode. Naturally, no reviewer has looked into it. At least we know it doesn't suck.

Stan

 

[Sean Keegan]

..are supposedly available or soon to be. I think the Nikon
adapter is available now and the Canon is coming soon.

What kind of adapter do you mean? Does it keep AF? What about AF-S lenses? What about lenses that do not have the aperture ring any more? Or only for certain lenses. Do you have a link? It sounds very interesting.

Regards,
Sean

 

[Rory]

so that I can crop to taste.

Rory

People argue that, constantly. Especially those who are proponents
of the "four thirds system".

The only people I know of that support a 4:3 system in the home display world are the uneducated masses who want all thier DVD's in "Full Screen" so they don't see those pesky black bars.. shrugs

I heard that a 4/3 aspect ratio is not necessarily part of the 4/3
standard.

Stan

--
Rory

 

[Stan]

What kind of adapter do you mean? Does it keep AF? What about AF-S lenses? What about lenses that do not have the aperture ring any more? Or only for certain lenses. Do you have a link? It sounds very interesting

Ya got me swinging. I read about this over on the Sigma forum. Maybe try a search to see if it's suitable or not for real world use. I think the Canon one is still being worked on.

Stan

 

[Sean Keegan]

No doubt, the SD9 is amazing for 3MP!!! Those look great. I constantly get results from my S2 that blow away any large enlargment I ever got from 35mm slide film in terms of noise and detail. I doubt film would be better than those SD9 examples either. Great post.

I also wear contacts, and when my wife takes photos of me with my camera I see my contacts a lot. Hehe.

Regards,
Sean

 

[Hamstor]

you ask for 'proof' yet you provide none yourself

the size is all relative, it doesn't matter if the pixel size is actually larger than the photodetector as Bayer is no different, so your point seems irrelevant

Don't forget, in Foveon, each pixel has a larger site area, so more
luminance is avaliable per site, even if each colour reduces the
avliable luminance to the next colour, it isn't as bad as you would
assume as there is more luminance to start with...

The size of the photosite is NOT necessarily the photo active area.
Indeed, almost all designs to date have significantly smaller photo
active areas than the photosite itself, so unless you can tell me
the actual area that reacts to light in a Foveon chip and have
compared that to other chips, please don't make such a claim.

As for the rest of your contention (more light available to the
third layer color), the math that I've seen doesn't support it.

--
Thom Hogan
author, Nikon Field Guide & Nikon Flash Guide
author, Complete Guides to the Nikon D100, D1, D1h, & D1x and
Fujifilm S2
http://www.bythom.com

 

[Hamstor]

ah, stop and think about it, of course RAW data has no colour! Each detector for each colour is the same type of detector, it doesn't detect colour, it detects light levels. Only by the fact there are colour filters obove the sensors does the post processing module (in camera or computer) convert the grey levels to a colour level. So of course in pure RAW data, there is actually no colour...

I know for me, for a long time I
assumed each detector actually detected RGB, so I still have to
conciously fight the subconcious conception of the way the bayers
actually work.

More or less the same way as human's eye.

Don't forget, in Foveon, each pixel has a larger site area, so more
luminance is avaliable per site, even if each colour reduces the
avliable luminance to the next colour, it isn't as bad as you would
assume as there is more luminance to start with...

Have you ever seen RAW linear data from Foveon chip? There is
nearly no color, it looks gray. If you have PS, try using Channel
Mixer on normal color image 40,30,30; 30,40,30; 30,30,40 - you
would see something very much like X3 output. Now try restoring
color on that image after Channel Mixer distortion. For details,
see foveon_interpolate routine in dcraw.

--
no text

 

[Ron Parr]

you ask for 'proof' yet you provide none yourself

the size is all relative, it doesn't matter if the pixel size is
actually larger than the photodetector as Bayer is no different, so
your point seems irrelevant

Pixels in APS CMOS sensors lose some of their sensing area to transistors, which reside on the surface of the chip. If you have more photodiodes per pixel, you need more transistors per pixel, which means that you sacrifice a larger fraction of the pixel area to transistors.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/