Bayer close to limit, Foveon just started?

this new sigma SD1 news is really exciting

as Bayer sensor gets close to 18m on APS-C size on a 2-d surface, is it close to physics limit? and it's wasting precious space using 4 spots to get 1 full color

on the other hand, Foveon is a 3-d sensor, each spot on the 2-d surface get you 1 full color, so we can expect it has 3x more room to grow while Bayer has hit the wall...

any thoughts?

ns66 said:

this new sigma SD1 news is really exciting

Click to expand...

Agreed.

ns66 said:

as Bayer sensor gets close to 18m on APS-C size on a 2-d surface, is it close to physics limit?

Click to expand...

Nope. The "physics limit", using resonant cells, is 6.6 gigapixels for an APS C sensor. But that's a little insane, since no lens can resolve that high.

The practical limit, below which we're below the diffraction limit at f1.4, is about 180mp on an APS-C sensor. That's 2.3 micron cells, which should more than exceed the noise, dynamic range, and high ISO capability of the Foveon that's just been announced, with 5 micron cells.

But that's not important, either.

ns66 said:

and it's wasting precious space using 4 spots to get 1 full color

Click to expand...

No, that's not how Bayer sensors work, at all. Most of the resolution comes from the green cells, which is half the pixels. The red and blue actually increase the resolution of that. This is why the previous generation Sigma (SD14, SD15) at 4.6mp were generally considered, when comparing prints at equal sizes, about equal to 10mp Bayer cameras.

ns66 said:

on the other hand, Foveon is a 3-d sensor, each spot on the 2-d surface get you 1 full color, so we can expect it has 3x more room to grow while Bayer has hit the wall...

Click to expand...

Not even close.

ns66 said:

any thoughts?

Click to expand...

• The real advantage of a Foveon sensor was its on-chip binning capability. It has the best, easiest to implement liveview and video modes of any sensor ever built. But that was 2002. In 2008, Canon launched Bayer sensors with horizontal binning, and this year we're going to see a lot of Bayers with both horizontal and vertical binning.
• As the sensor resolution exceeds the optical resolution of the lenses, the Bayer AA filter goes away (they've been getting weaker over the years). That puts a Foveon sensor at a data disadvantage to a Bayer sensor. The new 15mp sensor should match a 30mp Bayer resolution-wise, but it produces 50% more data, equivalent to 45mp.
• Bayer sensors don't have a lower limit on color accuracy, they can achieve literally perfect (100% match to human eye perception) color. Foveon sensors have a lower limit of 6% color errors due to failure of observer metamerism.

In short, this new camera is interesting because it dares to push the pixel count so high. No one has tried an equivalent 30mp Bayer yet. It's a novelty, not a revolution.

--
Rahon Klavanian 1912-2008.

Armenian genocide survivor, amazing cook, scrabble master, and loving grandmother. You will be missed.

Ciao! Joseph

http://www.swissarmyfork.com

another thought, right now with the best lens on a best bayer sensor, the per-pixel level sharpness is almost as good as it gets (well with some sharpening from raw), I wonder how much Foveon can do to improve the same image..., maybe the sharpness will come straight from raw with no PP...

from Foveon vs bayer regarding diffraction limits? lets say the same lens (a sigma 17-50mm for an example) on a bayer 30mp body and on the SD1 body....?....

ns66 said:

another thought, right now with the best lens on a best bayer sensor, the per-pixel level sharpness is almost as good as it gets (well with some sharpening from raw), I wonder how much Foveon can do to improve the same image..., maybe the sharpness will come straight from raw with no PP...

Click to expand...

--

Think about photography books and classes before any kind of gear....without knowledge, no camera is useful....

Carlos Roncatti Bomfim

As someone whose primary display media are a 1080p 46" TV and a 120" Panasonic 1080p front projector, what benefit if any would I be getting from the increased resolution of this camera over my D90? I have no urge to print any of my masterpieces larger than 16X20 at the outside.

thanks for the comment, the "physics limit" i actually mean the practical limit including everything that makes better image not practical or not worth it, governed by physics laws.

the color fidelity is an issue for Foveon sensor that sigma must resolve, I used to have a thought, that might be mathematically impossible, for a light of any color that hits the pixel surface, you get 3 readings, you need to get a function with constant parameters that computes the correct color/strength of the original light using that 3 numbers, no matter what spectrum of light it was, it's a magic by itself

carlos roncatti said:

from Foveon vs bayer regarding diffraction limits? lets say the same lens (a sigma 17-50mm for an example) on a bayer 30mp body and on the SD1 body....?....

Click to expand...

Why would they be much different? The "per pixel sharpness" and "apparent detail beyond nyquist" of the Foveon sensor will be just as affected by diffraction as the 30mp mosaic. You cannot rationally both claim the Foveon can resolve extra detai beyond the sensel countl and yet have that detail magically unaffected by diffraction (at least at the gross level.)

--
Erik

Reilly Diefenbach said:

what benefit if any would I be getting from the increased resolution of this camera over my D90?

Click to expand...

A similar benefit (or lack thereof) as you would get from a D3x (or hypothetical D700x or D400 with higher pixel count.) If you have what you need, then there is not much reason to change.

--
Erik

let me try an over over simplifed model just to illustrate:

original light with R/G/B strenth of r/g/b,
after hit layer 1, reduce strenth by fr1/fg1/fb1 respectively,
hit layer 2, reduce strenth by fr2/fg2/fb2,
hit layer 3, reduce strenth by fr3/fg3/fb3,

now 3 reading from senor

L1=fr1*r+fg1*g+fb1*b

L2=fr2*r+fg2*g+fb2*b

L3=fr3*r+fg3*g+fb3*b

now we can use different 3 sets of r/g/b values (input light) and resulting readings to solves the 9 parameters, but that's it, when a million different sets of r/g/b values (different color light) hit, these 3 equations with determined 9 parameters need to hold true!
imagine that

this might not be mathematical possible, that's why foveon is a struggle to get ALL the colors right

bayer is simpler since it works just like the monitor to combine r/g/b "flat" to deceive the eye and brain

well just a thought

also note when limited by lens resolving power, foveon has 3x (or at least twice if using green) more room to reach that from sensor side

Curious to hear your opinion on the value of the Foveon chip for black and white photography. I realize the issues with color accuracy (even so, I am happy with my SD14 for fine detail closeup work and panoramas), but shouldn't there be some value in capturing identically sampled luminance values at each pixel instead of capturing filtered values at adjacent pixels and then guessing at the actual luminance values as done with a Bayer array?

Cheers, Keith
http://www.kotay.net/keith/photo/photo.shtml

ns66 said:

let me try an over over simplifed model just to illustrate:

original light with R/G/B strenth of r/g/b,
after hit layer 1, reduce strenth by fr1/fg1/fb1 respectively,
hit layer 2, reduce strenth by fr2/fg2/fb2,
hit layer 3, reduce strenth by fr3/fg3/fb3,

now 3 reading from senor

L1=fr1*r+fg1*g+fb1*b

L2=fr2*r+fg2*g+fb2*b

L3=fr3*r+fg3*g+fb3*b

now we can use different 3 sets of r/g/b values (input light) and resulting readings to solves the 9 parameters, but that's it, when a million different sets of r/g/b values (different color light) hit, these 3 equations with determined 9 parameters need to hold true!
imagine that

this might not be mathematical possible, that's why foveon is a struggle to get ALL the colors right

bayer is simpler since it works just like the monitor to combine r/g/b "flat" to deceive the eye and brain

well just a thought

Click to expand...

A system of 3 linear equations with 3 unknowns is trivial to solve efficiently. No problem whatsoever for modern day chips.

GideonW said:

A system of 3 linear equations with 3 unknowns is trivial to solve efficiently. No problem whatsoever for modern day chips.

Click to expand...

i think you missed my point, i am not talking about calculation speed or anything like that, but about color shift on foveon sensor.

to get color fidelity for ALL colors, that means foveon need to pick the 9 parameters (and burn inside the foveon ROM) to make the 3 equations equal no matter what input light is, it's proven that doesn't work well, that's the reason of the famous foveon color shift, since they can get some color right by tuning the 9 parameters, but other colors will be off

UnderDriven said:

Curious to hear your opinion on the value of the Foveon chip for black and white photography. I realize the issues with color accuracy (even so, I am happy with my SD14 for fine detail closeup work and panoramas), but shouldn't there be some value in capturing identically sampled luminance values at each pixel instead of capturing filtered values at adjacent pixels and then guessing at the actual luminance values as done with a Bayer array?

Cheers, Keith
http://www.kotay.net/keith/photo/photo.shtml

Click to expand...

there could be some value, but to be honest I am not sure that's biggest advantage foveon provides, also it very much depends on algorithms used to calculate luminance, so...

Erik Magnuson said:

carlos roncatti said:

from Foveon vs bayer regarding diffraction limits? lets say the same lens (a sigma 17-50mm for an example) on a bayer 30mp body and on the SD1 body....?....

Click to expand...

Why would they be much different? The "per pixel sharpness" and "apparent detail beyond nyquist" of the Foveon sensor will be just as affected by diffraction as the 30mp mosaic. You cannot rationally both claim the Foveon can resolve extra detai beyond the sensel countl and yet have that detail magically unaffected by diffraction (at least at the gross level.)

Click to expand...

But ... if the hypothetical 30 Mp Bayer sensor had the exact same size as the SD1 sensor , the Bayer sensor would have a much higher pixel -density = smaller pixels = more impacted by diffraction .. Wouldn´t it ??

Member said:

--
Erik

Click to expand...

--
Please visit my galleries at :
http://www.flickr.com/photos/yoicz/
or
http://www.naturephotos.dk/NaturePhotos_ejergalleri.php?menu=3&_Ejer=57

Joseph S Wisniewski said:

• Bayer sensors don't have a lower limit on color accuracy, they can achieve literally perfect (100% match to human eye perception) color. Foveon sensors have a lower limit of 6% color errors due to failure of observer metamerism.

Click to expand...

You keep quoting those numbers. Any references ? I mean serious one, not DP forum messages.

Joseph S Wisniewski said:

In short, this new camera is interesting because it dares to push the pixel count so high. No one has tried an equivalent 30mp Bayer yet. It's a novelty, not a revolution.

--
Rahon Klavanian 1912-2008.

Armenian genocide survivor, amazing cook, scrabble master, and loving grandmother. You will be missed.

Ciao! Joseph

http://www.swissarmyfork.com

Click to expand...

--
http://www.flickr.com/photos/maximebrousse/

Joseph S Wisniewski wrote:
.

Joseph S Wisniewski said:

• As the sensor resolution exceeds the optical resolution of the lenses, the Bayer AA filter goes away (they've been getting weaker over the years). That puts a Foveon sensor at a data disadvantage to a Bayer sensor. The new 15mp sensor should match a 30mp Bayer resolution-wise, but it produces 50% more data, equivalent to 45mp.

Click to expand...

That's the raw file. But when you start working with the image data, you are working with a 15 MP image file that has as much detail as the 30MP bayer image output. So for intermediate work in Photoshop, the Foveon files are more compact.

If for some reason you chose to shoot only JPG, the foveon sensor would again have an advantage offering more detail in a smaller container.

Joseph S Wisniewski said:

• Bayer sensors don't have a lower limit on color accuracy, they can achieve literally perfect (100% match to human eye perception) color. Foveon sensors have a lower limit of 6% color errors due to failure of observer metamerism.

Click to expand...

100% match?

Yes, if pointed at a subject of solid color with no edges.

The problem is the real world is full of examples where color meets color. And then because those samples which are totally split in terms of spectrum giving you that perfect accuracy, are not as accurate for any given output pixel because they are not sampling the same data. That's why you get repeated color errors across generations of bayer cameras like these examples from the M9 and the 23MP D3x that show exactly the same issues:

http://www.pbase.com/kgelner/image/117316541



http://www.pbase.com/kgelner/image/128681825



In both cases you get a red box changing the color of the railing in front of it. And you get the fun effect of having a joggers legs in front of the railing the jogger is actually behind, because the bars turn flesh (or pant) colored!

These are not pixel-peeping issues. These are effects you can see plainly when the whole image is presented in a browser, that you could see on a 5x7 print never mind a larger one.

100% accuracy for a bayer sensor is only theoretically possible where the subject is one single color across the frame. In any other case displacement of the samples WILL introduce some error. On average I'm sure it's slight, but it's not zero and I don't think in real conditions is better than the Foveon chips fare.

--
---> Kendall
http://InsideAperture.com
http://www.pbase.com/kgelner
http://www.pbase.com/sigmadslr/user_home

UnderDriven said:

Curious to hear your opinion on the value of the Foveon chip for black and white photography. I realize the issues with color accuracy (even so, I am happy with my SD14 for fine detail closeup work and panoramas), but shouldn't there be some value in capturing identically sampled luminance values at each pixel instead of capturing filtered values at adjacent pixels and then guessing at the actual luminance values as done with a Bayer array?

Click to expand...

Yes, you should get smoother and more accurate tonal transitions.

And you have full control over adjusting the color channel mix for the final B&W result. Some people remove the CFA to get a monochrome only bayer sensor with a ton of resolution, which is find but then you have locked in your mix to treat all colors equally for tone.

I like the Foveon cameras for tone, but they also have great DR for B&W photography...

--
---> Kendall
http://InsideAperture.com
http://www.pbase.com/kgelner
http://www.pbase.com/sigmadslr/user_home

Reilly Diefenbach said:

As someone whose primary display media are a 1080p 46" TV and a 120" Panasonic 1080p front projector, what benefit if any would I be getting from the increased resolution of this camera over my D90? I have no urge to print any of my masterpieces larger than 16X20 at the outside.

Click to expand...

If you only displayed your images at 100% on the TV, you will derive zero benefit from the resolution increases achieved in the past few years. A 1080p TV has a resolution of 1920x1080 pixels, which equals 2 megapixels.

Two megapixels.

You could fill up your TV's resolution with a phone camera (not accounting for what the lens can actually resolve). A D90 is already serious overkill for an HDTV image. That's partly why DSLRs can so easily handle shooting HDTV video: the resolution HDTV are so much smaller than what the sensor is capable of, that shooting 30 frames per second of HDTV is no problem at all.

Of course, if you like to be able to zoom in on the images on your TV, you can use a lot more than 2 megapixels.

16x20 prints are a different matter, based on the optimal resolution of your printer, inks, and paper. If your printer can print meaningful detail up to 300 dpi, then ideally, a file for a 16x20 print should be 4800x6000 pixels, or around 29 megapixels. But that's assuming you're using the highest quality setting, careful expert processing, and the best glossy paper, with an image taken with a sharp lens and zero motion blur. With lesser equipment and technique, lower resolutions (12 megapixels and up) will look just as good.

In my opinion, the Foveon technology will run into the resolution before the Bayer technology does. This is because Foveon read red from a relatively deep sensor. The light will be diffuse by the time it gets there earlier for a Foveon than a Bayer sensor. It's probably not that significant of a difference, though.
--