FoVeOn X3 vs. Canon 6MP CMOS

[richard kim]

so i just saw the sample images of the x3 3.3 megapixel chip, and i have to say i'm freakin amazed....

is the canon cmos 6 mega going to be comparable or would it be better to get the x3 even though it's 3.3 mega.....

ugh i don't know what to do....
it's kinda like asking

1 Gig G4 apple (x3 foveon chip)
or
2.2 P4 PC (6MB CMOS).....

anyone have any true technical insight? -- there is no spoon~~

 

[Nicholas81360]

Saw? What? Where? Please share. Before I buy D60 would like to get lowdown on X3. I'm a D30 photog right now.

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....
--The joy of photography is being there when you take the picture.

 

[orlandoom]

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....

--
The joy of photography is being there when you take the picture.

I don't know if this is what he saw but I found this link.
http://www.sjphoto.com/web-special/index.htm

 

[rob evans]

Yeah. We are all waiting for real samples from the 3.3 megapixel chip. The site given is apparently not this chip. Secondly, we all want to see real dslr functionality and performance.

Its hard to buy anything with this thing hanging over our heads though isn't it. I sure hope this isn't smoke.

rob evans

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....

--
The joy of photography is being there when you take the picture.

I don't know if this is what he saw but I found this link.
http://www.sjphoto.com/web-special/index.htm

 

[budhines]

Hi All; It's sort of like Celera Geneomics (sp) CRA. The human gene whiz kid who was all over forbes, etc. covers last year. He let it all slip through his fingers and his stock went down by 90%! Most of these guys are not good businessmen. I would say we have a one in five shot of this chip seeing the light of day, at least in any form thats useful for more than just point and shoots. Of course that would make him a good businessman wouldn't it! It's sure nice to think about though. Remember digital film? The one you just snapped in the back of your camera? It was coming out any day now for years.

Bud

Its hard to buy anything with this thing hanging over our heads
though isn't it. I sure hope this isn't smoke.

rob evans

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....

--
The joy of photography is being there when you take the picture.

I don't know if this is what he saw but I found this link.
http://www.sjphoto.com/web-special/index.htm

 

[Samir Kharusi]

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....
is the canon cmos 6 mega going to be comparable or would it be
better to get the x3 even though it's 3.3 mega.....

Simple physics and signal modulation theory (Nyquist Sampling Theorem) says that Foveon sensors with 10 micron square pixels should be able to resolve 50 line-pairs per mm. If you check out your D30 today (I did a few weeks back) you'll find that its 10 micron pixels are able to resolve 43 lp/mm vertically and horizontally and 48 lp/mm along the diagonal. This is despite the Bayer pattern and the endless talk disparaging it. So the difference between Bayer and Foveon sensors is not as huge as Foveon hype often implies. In fact by analogy from the D30 one can say that the D60 sensor, 7.4 micron pixels, should be able to resolve somewhere between 55 and 60 lp/mm horizontally and vertically. I have not yet laid my hands on a D60 but from Phil's review it looks like that's exactly what is happening. 55 to 60 lp/mm = 1660 to 1800 lines per picture height. On Phil's test shots you have to be aware that the lenses are not always up to the job, so just check out his best samples at f8. Yes, the Foveon is revolutionary in concept but not necessarily revolutionary in delivery. That will depend on how easy it is to fabricate, its quantum efficiency, noise, etc, etc. Rest easy and get your new toy now. Whatever happens to Foveon, your D60 will definitely be passe in 18 months, but that's digital life :-(

 

[Z (is real)]

Simple physics and signal modulation theory (Nyquist Sampling
Theorem) says that Foveon sensors with 10 micron square pixels
should be able to resolve 50 line-pairs per mm. If you check out
your D30 today (I did a few weeks back) you'll find that its 10
micron pixels are able to resolve 43 lp/mm vertically and
horizontally and 48 lp/mm along the diagonal. This is despite the
Bayer pattern and the endless talk disparaging it.

Not for scenes of interesting color complexity. (Though we can argue all day about the definition of "resolve". Suffice it to say that color artifacts are noticeable in resolution chart images for most digital SLRs and are often noticeable in areas of fine detail in real scenes.)

The main advantage to Foveon X3 will not be in measured resolution, but in the lack of certain artifacts in its output. As such, it will judged by its images as are all other digital cameras. It is an interesting technology, though I doubt it will be doing much in the digital SLR space for the next year. (I.e. I doubt many people looking at D60/D100 class cameras, much less EOS-1D/D1x type bodies, will be interested in the Sigma SD9.)

-Z-

 

[Forrest41926]

The main advantage to Foveon X3 will not be in measured resolution,
but in the lack of certain artifacts in its output. As such, it
will judged by its images as are all other digital cameras. It is
an interesting technology, though I doubt it will be doing much in
the digital SLR space for the next year. (I.e. I doubt many people
looking at D60/D100 class cameras, much less EOS-1D/D1x type
bodies, will be interested in the Sigma SD9.)

I think it's interesting ... and maybe even worth the trade-off of who-knows how bad autofocus and metering ... but not worth the lens mount. In a year or two I'd like to see the technology in a Canon body, though.

 

[JoePhoto]

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....
is the canon cmos 6 mega going to be comparable or would it be
better to get the x3 even though it's 3.3 mega.....

Simple physics and signal modulation theory (Nyquist Sampling
Theorem) says that Foveon sensors with 10 micron square pixels
should be able to resolve 50 line-pairs per mm. If you check out
your D30 today (I did a few weeks back) you'll find that its 10
micron pixels are able to resolve 43 lp/mm vertically and
horizontally and 48 lp/mm along the diagonal. This is despite the
Bayer pattern and the endless talk disparaging it. So the
difference between Bayer and Foveon sensors is not as huge as
Foveon hype often implies. In fact by analogy from the D30 one can
say that the D60 sensor, 7.4 micron pixels, should be able to
resolve somewhere between 55 and 60 lp/mm horizontally and
vertically. I have not yet laid my hands on a D60 but from Phil's
review it looks like that's exactly what is happening. 55 to 60
lp/mm = 1660 to 1800 lines per picture height. On Phil's test shots
you have to be aware that the lenses are not always up to the job,
so just check out his best samples at f8. Yes, the Foveon is
revolutionary in concept but not necessarily revolutionary in
delivery. That will depend on how easy it is to fabricate, its
quantum efficiency, noise, etc, etc. Rest easy and get your new toy
now. Whatever happens to Foveon, your D60 will definitely be passe
in 18 months, but that's digital life :-(

Maybe I am misunderstanding your post ... but it seems like you are missing the point that it REQUIRES 3 PIXELS, (rgb), to produce "ONE" pixel of "white" light. Thus when you do your size calculations/comparasons .. you actually have to measure the groups of three.

Phils sample images had very bad pixelazation ... particularly in the reds.

Digital does NOT meet my standards yet.

 

[Samir Kharusi]

Maybe I am misunderstanding your post ... but it seems like you are
missing the point that it REQUIRES 3 PIXELS, (rgb), to produce
"ONE" pixel of "white" light. Thus when you do your size
calculations/comparasons .. you actually have to measure the groups
of three.

Phils sample images had very bad pixelazation ... particularly in
the reds.

Digital does NOT meet my standards yet.

Indeed perhaps we are not understanding each other. Using Bayer interpolation you do NOT need 3 pixels to equal one Foveon pixel, as some of the clever wording implies in Foveon statements. Foveon uses wording that is technically sound to those who understand how Bayer interpolation works but is misleading to the layman. Resolution is primarily determined by luminosity. Your eyes are much more critical of luminosity changes than they are of colour changes. If the luminosity channel is sharp then the colour channels can be very unsharp and the image will still appear sharp. This is a fundamental principle that is in daily use for all colour television broadcasting and is also fundamental to the LRGB process that is in routine use by amateur astronomers. My own tests clearly indicate that Moire fringes, colour smears, etc are bothersome mainly at > 40 lp/mm in the D30. In a Foveon with same size pixels but no Bayer filters you should be able to go all the way to 50 lp/mm before resolution becomes zilch. So the difference is between 40+ and 50, not 40+ and 120 as Foveon's rather clever wording implies. Yes the Bayer method does require colour info from surrounding pixels but the resolution after processing is pretty close to that given by a single pixel. Sounds magical but it ain't; just clever. What's even more amazing is that it can all be done with simple arithmetic using RGB values of the surrounding pixels. By the way, the appearance of square pixels in images is also trivially easy to get rid of. Upsizing with standard bicubic interpolation in any image editor makes it disappear instantly. The image may still not be as sharp as you would like, which is perhaps your reason to stick to film (especially medium format), but the pixels can easily be smoothed out and made not to show. I presume Phil may not have bothered since he was specifically trying to demonstrate jagged diagonals to a digital-aware audience or whatever in the examples you cite. While pixels can be smoothed out without defocusing the digital image, unfortunately one cannot smooth out grain as easily without mushing up the image resolution ;-)

 

[JoePhoto]

Maybe I am misunderstanding your post ... but it seems like you are
missing the point that it REQUIRES 3 PIXELS, (rgb), to produce
"ONE" pixel of "white" light. Thus when you do your size
calculations/comparasons .. you actually have to measure the groups
of three.

Phils sample images had very bad pixelazation ... particularly in
the reds.

Digital does NOT meet my standards yet.

Indeed perhaps we are not understanding each other. Using Bayer
interpolation you do NOT need 3 pixels to equal one Foveon pixel,
as some of the clever wording implies in Foveon statements. Foveon
uses wording that is technically sound to those who understand how
Bayer interpolation works but is misleading to the layman.
Resolution is primarily determined by luminosity. Your eyes are
much more critical of luminosity changes than they are of colour
changes. If the luminosity channel is sharp then the colour
channels can be very unsharp and the image will still appear sharp.
This is a fundamental principle that is in daily use for all colour
television broadcasting and is also fundamental to the LRGB process
that is in routine use by amateur astronomers. My own tests clearly
indicate that Moire fringes, colour smears, etc are bothersome
mainly at > 40 lp/mm in the D30. In a Foveon with same size pixels
but no Bayer filters you should be able to go all the way to 50
lp/mm before resolution becomes zilch. So the difference is between
40+ and 50, not 40+ and 120 as Foveon's rather clever wording
implies. Yes the Bayer method does require colour info from
surrounding pixels but the resolution after processing is pretty
close to that given by a single pixel. Sounds magical but it ain't;
just clever. What's even more amazing is that it can all be done
with simple arithmetic using RGB values of the surrounding pixels.
By the way, the appearance of square pixels in images is also
trivially easy to get rid of. Upsizing with standard bicubic
interpolation in any image editor makes it disappear instantly. The
image may still not be as sharp as you would like, which is perhaps
your reason to stick to film (especially medium format), but the
pixels can easily be smoothed out and made not to show. I presume
Phil may not have bothered since he was specifically trying to
demonstrate jagged diagonals to a digital-aware audience or
whatever in the examples you cite. While pixels can be smoothed out
without defocusing the digital image, unfortunately one cannot
smooth out grain as easily without mushing up the image resolution
;-)

Thank You for that very coherent response. I first of all concede that I know nothing about Bayer yet ... that is one of the reasons I am here. As a pro for 40 years ... and previous owner of a photo LAB; I originally saw digital as a threat that eventually contributed to my closing my lab; (90% of my business was comercial Real Estate Appraisers or Insurance, etc... and they were the first to go digital).

But I am also an Electronics Enginner so I do "understand" digital.

I would like to find out more about "Bayer" ... but it still seems like a form of "interpolation"; (albeit maybe the best form so far). And while I appreciate it can eliminate the "pixel" problem on any size print ... it still has no "more" data than in the original. The "additional" information is still a "guess" and can therefore be wrong as many times as it is right.

It just seems to me like the X3 has at least the potential of providing 3X more "original" data ... and therefore even further interpolation to get even bigger sizes would be that much more accurate.

BUT; I still concede that we need to WAIT to see the RESULTS. There may still be other negatives, (noise, sensitivity and dynamic range, etc. ) that we don't know yet.

 

[Harold Burns]

I would say we have a
one in five shot of this chip seeing the light of day, at least in
any form thats useful for more than just point and shoots.

Foveon has claimed to have the next best thing in chip technology for the last several years, but with each claim they have had a lot more success generating media coverage than with delivering products that photographers want to buy.

Harold Burns

 

[Michael OHara]

The main advantage to Foveon X3 will not be in measured resolution,
but in the lack of certain artifacts in its output. As such, it
will judged by its images as are all other digital cameras. It is
an interesting technology, though I doubt it will be doing much in
the digital SLR space for the next year. (I.e. I doubt many people
looking at D60/D100 class cameras, much less EOS-1D/D1x type
bodies, will be interested in the Sigma SD9.)

I think it's interesting ... and maybe even worth the trade-off of
who-knows how bad autofocus and metering ... but not worth the lens
mount. In a year or two I'd like to see the technology in a Canon
body, though.

What you said, but make that a Nikon body too please!!--Michael OHara / WetPlanet / Honoluluwww.DiveSlates.com

 

[Erik Magnuson]

I would like to find out more about "Bayer" ... but it still seems
like a form of "interpolation"; (albeit maybe the best form so
far).

Here are some links:

http://www.fillfactory.com/htm/technology/htm/rgbfaq.htm

For more on different algorithms, try:

http://www-ise.stanford.edu/~tingchen/intro.htm

but this link is a bit unreliable.
--ErikFree Windows JPEG comment editor http://home.cfl.rr.com/maderik/edjpgcom

 

[ramilappalainen]

It´s interesting to read your guys comments about the resolving quality of the sensors. I really apreciate your knowledge about these digital systems!

I own Canon D30 and have done some testes about how sharp iamges my optics can produce. The only one which produces sharp images is 50/1.4 Other 70-200/2.8 and old 20-35/2.8 are actually quite bad in terms of sharpness

The question is: is there any links about canon lenses and their resolving power? And have I understand right that if my D30 resolving power is about 40+ lpmm my optics should give at least the same factor that the result is sharp image?

I don´t even know what is the resolving power of my 50/1.4 but I have seen the difference when shooting to film...

 

[E. Blues]

so i just saw the sample images of the x3 3.3 megapixel chip, and i
have to say i'm freakin amazed....
is the canon cmos 6 mega going to be comparable or would it be
better to get the x3 even though it's 3.3 mega.....

I confess that I would not know an RGB from an MGB, or a Bayer from a Tylenol, but I have to wonder why, if the Foveon technology is so superior, it's only available in a camera almost nobody can afford, (Foveon) or a camera almost nobody wants, (Sigma). This is assuming Sigma actually produces such a camera. Doesn't anybody wonder why they were unable to peddle this thing to Nikon or Canon?--Just my opinion, and you know what opinions are like. Everybody has one.EB

 

[Erik Magnuson]

The question is: is there any links about canon lenses and their
resolving power?

Try http://www.photodo.com/nav/prodindex.html--ErikFree Windows JPEG comment editor http://home.cfl.rr.com/maderik/edjpgcom

 

[Erik Magnuson]

Doesn't anybody
wonder why they were unable to peddle this thing to Nikon or Canon?

It's all speculation, but some rumours have emerged:

Sigma wanted to enter the digital market badly. Rather than just introduce a me-too camera based on a standard CCD, they wanted something that would enable them to make their name in digital. So they were willing to commit the resources to get a camera out by June/July. If the rumours of a 2yr exclusive are true, they may have been willing to pay a premium.

Fuji, Canon and Nikon have existing DSLR lines and development plans. The first generation of X3 sensors is likely to have some limitations (1.7x multiplier, ISO

Rather than Canon and Nikon, the more interesting question is did Minolta and/or Pentax pass or did Sigma just outbid them.--ErikFree Windows JPEG comment editor http://home.cfl.rr.com/maderik/edjpgcom

 

[Kenn Hwang60633]

Hi Samir,

One point I'd like to make is this: If all we're talking about is DPReview's black and white resolution chart, then any standard CCD will be able to reproduce it at nearly full resolution. At that point it falls to the number, density, and distribution of each color sensor on the CCD/CMOS regardless of its color filter (yes, the lens resolution matters as well). All the filters will report true gray with similar/identical values, so the "bayar" issue is nonexistent.

However, if you had the same resolution chart in deeply saturated, red, blue, or to a lesser extent green, you'll see a big difference in the result a Foveon-type sensor would render the image. Depending on how "pure" and highly saturated the colors are, the bayar sensor really may deliver 1/4 the resolution of the X3. Of course, synthetic tests such as this (and that includes the black on white color charts) tend to expose the extremes of the bayar system (best case for b/w, worst-case for primary colors).

In real life, the situation varies drastically, and I would be hard-pressed to expect the Foveon to deliver anywhere near 3X the accuracy of a Bayar CCD. However, it will be tremendously more consistent with resolution accuracy under any circumstance (real-world or synthetic), and imo the samples we've seen so far are vastly superior to all but 3-CCD systems.

Of course, small Foveon sensors with tiny pixel-sensors will be just as prone to color-filter and luminosity noise as today's CCDs, and I expect the difference will be close to what we see now between the output of an CP990 and a D1.

Cheers,
Kenn

Indeed perhaps we are not understanding each other. Using Bayer
interpolation you do NOT need 3 pixels to equal one Foveon pixel,
as some of the clever wording implies in Foveon statements. Foveon
uses wording that is technically sound to those who understand how
Bayer interpolation works but is misleading to the layman.
Resolution is primarily determined by luminosity. Your eyes are
much more critical of luminosity changes than they are of colour
changes. If the luminosity channel is sharp then the colour
channels can be very unsharp and the image will still appear sharp.
This is a fundamental principle that is in daily use for all colour
television broadcasting and is also fundamental to the LRGB process
that is in routine use by amateur astronomers. My own tests clearly
indicate that Moire fringes, colour smears, etc are bothersome
mainly at > 40 lp/mm in the D30. In a Foveon with same size pixels
but no Bayer filters you should be able to go all the way to 50
lp/mm before resolution becomes zilch. So the difference is between
40+ and 50, not 40+ and 120 as Foveon's rather clever wording
implies. Yes the Bayer method does require colour info from
surrounding pixels but the resolution after processing is pretty
close to that given by a single pixel. Sounds magical but it ain't;
just clever. What's even more amazing is that it can all be done
with simple arithmetic using RGB values of the surrounding pixels.
By the way, the appearance of square pixels in images is also
trivially easy to get rid of. Upsizing with standard bicubic
interpolation in any image editor makes it disappear instantly. The
image may still not be as sharp as you would like, which is perhaps
your reason to stick to film (especially medium format), but the
pixels can easily be smoothed out and made not to show. I presume
Phil may not have bothered since he was specifically trying to
demonstrate jagged diagonals to a digital-aware audience or
whatever in the examples you cite. While pixels can be smoothed out
without defocusing the digital image, unfortunately one cannot
smooth out grain as easily without mushing up the image resolution
;-)

 

[S Clark]

Kodak is supposed to be looking into using the Foveon sensor-- http://www.pbase.com/galleries/sasc