Foveon news article is complete nonsense!

Berg Na said:

There will be a tremendous amount of crosstalk between
the different colors.

Click to expand...

This factor concerns me a lot too, but from a different angle.
But maybe we are talking about the same thing.

If you look at the spectral curves in the Merrill '875 patent,
you would see very broad spectral curves for each sensor.
In the tristimulus model, this is not, per se, a problem. You
take the area under each curve, you get three numbers.
You apply the proper transform and get xyz and you can get
anything else from there (ok, so that is an oversimplification...).
As long as the curves are not coincident (you know, the vectors
span the space nonsense we learned long ago in school) and
you know the curves perfectly accurately, you can, on paper,
get the result you want.

But... how closely do the spectral curves track from a sensor
taken from one wafer compared to the next? I am curious.

I also wonder what the suseptability to noise would do to a color
estimate, since the curves do overlap to such a large degree. Could
this lead to potential larger color error compared to a Bayer sensor,
especially in estimating the hue and saturation axes? So far, the
majority of the quantitative discussion on the Foveon was on resolution.
That one is easier to take a stab on. With this one, you have to guess
how much noise there is, and how consistent are the curves from sensor
to sensor; those of us who are free to discuss it don't have a clue .

For what I know, this may be non-problems to Foveon scientists and
engineers. The readout noise may be so low, it is not a problem at all.
Ditto the consistency. Perhaps the production chips have much sharper
spectral curves, who knows?

At the least, I would like to see the MacBeth chart taken at different
ISO numbers. I would hope to also that Phil can get two randomly selected
cameras to test consistency. (I can just hear him screaming: "two?! I
can't even get one bloody camera!"

• kc

Sigma 17~35 €600 variable aperture and prone to flair and I had to return one for dust inside the lens. A lens that a heavy user will wear out in a year or two.

Nikkor 17~35 €2,000 fixed aperture and closer focusing, tighter flair control and very solidly built. A lens that will last and last.

Good shooting weather and average subject matter and both lenses will perform above average – it is a personal decision if this is good enough for the range of photography one does. Certainly the price difference makes replacement every 12 months or so a viable alternative in favour of the Sigma.

I also plan on buying the D100, but only if it's

bill pitcher said:

sealed like the F100, not the F80. $2000+ is a way too much to pay
for a "sunny" day camera.

Click to expand...

Well that's all it is.

Even the D1x does not have the weather seals of the F100 film camera. The D1x will stop working in the rain - heavy rain and continued exposure but it will stop. It so far has come back each time though.

At $2,000 this is almost a disposable SLR in digital terms.

In the preview Phil said the test Foveon sensor was larger VERTICALLY, those extra pixels should not make any difference in resolving HORIZONTAL resolution, which is what the shot of VERTICAL lines implies

Michael

Gary Eickmeier said:

Phil Askey said:

I didn't post it for comparison, go back and read the article.

Click to expand...

No, you didn't post it for comparison in your article, but you are
using it for comparison in the text below:

Gary Eickmeier said:

Gary Eickmeier said:

Phil Askey said:

Correct, as you point out the "purity of the line" is important and
has a direct effect on output resolution. Did you see the
resolution chart crop I posted in my X3 preview? The Foveon X3 is
capable of resolving more monochrome resolution than an identical
MP bayer chip. And just as importantly the resolution chart could
use red, blue, orange, purple or any coloured line and you'd get
the same resolution from the Foveon.. Try that with a mosaic chip.

Click to expand...

Click to expand...

Click to expand...

Gary Eickmeier

Click to expand...

--Michael

Chris Brown said:

Member said:

The bigger point here I think is that a traditional Mosaic CCD is
undersampled by a factor of two in the green and three in the red
and blue channels. So the equivalent sampling of a 3.1 Mos CCD is
actually at best only 1.55 MP of real info (or 1.2 average RGB).

Click to expand...

This is misleading - Bayer sensors can actually recover image
information quite close to their pixel resolution, as the example
below (D30) illustrates:

Click to expand...

Sorry but missing information is missing information. No matter how you slice it, or interp it in this case, you're still making up information. It's always better to have real information when you upsample.

Your image doesn't prove anything. single pixel variations on a noisy sensor are even higher, but we know that if you interpolate it higher, it doesn't improve.--Photos, tips and tests at: http://www.geocities.com/glowluzid

Perhaps Trevor Bayliss can work on a clockwork digicam...

Ger Bee said:

Ger Bee said:

sealed like the F100, not the F80. $2000+ is a way too much to pay
for a "sunny" day camera.

Click to expand...

Well that's all it is.

Even the D1x does not have the weather seals of the F100 film
camera. The D1x will stop working in the rain - heavy rain and
continued exposure but it will stop. It so far has come back each
time though.

At $2,000 this is almost a disposable SLR in digital terms.

Click to expand...

jkirk@oz said:

Issue_1 : A 3.4 MP Faveon and a 3.4 MP Bayer Pattern have the SAME
RESOLUTION. ie If you are taking a shot of a very fine grid-pattern
(Phils V and H Res charts), they both are "capable" of resolving
the same number of "Lines". . . .

Click to expand...

jkirk@oz said:

jkirk@oz said:

Wrong. Bayer interpolation yields substantially less monochrome
resolution than is implied by the pixel count. . . .

Click to expand...

Click to expand...

jkirk@oz said:

Given the same "number of sites" (and layout) they both
"THEORETICALLY" can address the same V and H lines. Take out the
RGB lenses on Bayer and this is the case (assuming you also NOW
extract luminence from each site as well).

Click to expand...

Sort of. But taking the RGB filters away, and you aren't talking about Bayer at all. As long as we have color filters in a Bayer pattern, the luminance data is substantially worse than would be the case in a system where sensors and output pixels were related one-for-one without being averaged and interpolated. I think you're saying that a Bayer system would have higher resolution if it were something else. I guess we can agree on that.

jkirk@oz said:

If you bothered to read on, I then gave Bayer a "purity" de-rating. It
couldn't be any clearer.

Click to expand...

Whatever.

Berg Na said:

Berg Na wrote:
The color signals in the Foveon chip will be anything but "pure" or
"perfect". There will be a tremendous amount of crosstalk between
the different colors...

Click to expand...

Now we're in the realm of "SPECULATION". The "purity of line" is Empirical. Both systems will have "noise". Bayer has had years to improve this but it STILL remains a major issue. 'Empirically" the Foveon model can reach a far greater "purity" than the Bayer model.

Berg Na said:

.....There are many factors that affect the
absorption depth in silicon, i.e., local crystal quality and doping
concentration, so separating the different colors as a function of
thickness requires a fancy algorithm that is no different in
complexity compared to the bayer color interpolation method.

Click to expand...

Except "ALL" your information is at the same site, you don't have to go "begging" the rest (most) of the info from your neighbours.

Berg Na said:

Foveon's simple explanation of the device operation is far from
complete.

Click to expand...

Berg Na said:

jkirk@aussie said:

... The "purity of the Line" would be far superior with the Faveon
though, because it can "resolve" perfect colour at eack "pixel" or
"line element". In the bayer, you only have a red then blue then
green / green pixel, so if it was a "red line" you'd miss
resolution every 4 !! You can't MAKE UP what it was supposed to be
!!!

Click to expand...

Click to expand...

There is no need to speculate, the physics of light absorption in silicon are rather simple. Empirically? What experimental data do you have? The Foveon approach will generate a lot of intermixing between the different colors, nothing close to 'purity'.

jkirk@aussie said:

Now we're in the realm of "SPECULATION". The "purity of line" is
Empirical. Both systems will have "noise". Bayer has had years to
improve this but it STILL remains a major issue. 'Empirically" the
Foveon model can reach a far greater "purity" than the Bayer model.

Click to expand...

The problem of signal contamination is actually much worse than a limited signal.

jkirk@oz said:

Except "ALL" your information is at the same site, you don't have
to go "begging" the rest (most) of the info from your neighbours.

Click to expand...

Michael w. said:

In the preview Phil said the test Foveon sensor was larger
VERTICALLY, those extra pixels should not make any difference in
resolving HORIZONTAL resolution, which is what the shot of VERTICAL
lines implies

Click to expand...

My understanding of lines per picture height res tests is that you size up the resolution chart in your viewfinder by making it exactly fill the finder from top to bottom. So, for a square imager, you would be zooming in more on the test chart than for a rectangular imager of the same width. I could be mistaken on this, but if I'm right, this means that both vertical and horizontal resolution have an advantage (seeming advantage) with the square chip.

Gary Eickmeier

You can think of it as if you are using the square part of a larger sensor.

So, for a 2K by 2K sensor, you're using the square part of 2k by 3K = 6MP sensor.

--Ron ParrFAQ: http://www.cs.duke.edu/~parr/photography/faq.htmlGallery: http://www.pbase.com/parr/

Yes, that's spot on, Gary, the resolution on the chart has to be discounted for the fact that the tested sensor's height (2048 pixels) is greater than that of the SD9 (1512 pixels). If you look at the chart, you can see that the lines are clearly resolved out to the end of the chart (2000 l/ph) so we don't really know how much further they'd go (new chart required for the latest chips!) so the ultimate resolution of the SD9's sensor is still a bit of a mystery, but I think it's safe to say it will be comparable with the D60, S2 and D100 (around 1600 l/ph). In fact we are probably moving quite quickly into an era where all the top-line chips have plenty of resolution for all normal purposes and other considerations will become more significant to image quality.

Gary Eickmeier said:

Michael w. said:

In the preview Phil said the test Foveon sensor was larger
VERTICALLY, those extra pixels should not make any difference in
resolving HORIZONTAL resolution, which is what the shot of VERTICAL
lines implies

Click to expand...

My understanding of lines per picture height res tests is that you
size up the resolution chart in your viewfinder by making it
exactly fill the finder from top to bottom. So, for a square
imager, you would be zooming in more on the test chart than for a
rectangular imager of the same width. I could be mistaken on this,
but if I'm right, this means that both vertical and horizontal
resolution have an advantage (seeming advantage) with the square
chip.

Gary Eickmeier

Click to expand...

--Richard C. South Australia

Ron Parr said:

You can think of it as if you are using the square part of a larger
sensor.

So, for a 2K by 2K sensor, you're using the square part of 2k by 3K
= 6MP sensor.

Click to expand...

No.

The Sigma sensor is about 1512 x 2268 pixels. This means for the rectangular sensor, you are using only 1512 pixels vs 2048 with the sqare imager.

Or are you comparing the square Foveon with the 6mp Canon?

Gary Eickmeier

Gary Eickmeier said:

Ron Parr said:

You can think of it as if you are using the square part of a larger
sensor.

So, for a 2K by 2K sensor, you're using the square part of 2k by 3K
= 6MP sensor.

Click to expand...

No.

Click to expand...

Yes!

Gary Eickmeier said:

The Sigma sensor is about 1512 x 2268 pixels. This means for the
rectangular sensor, you are using only 1512 pixels vs 2048 with the
sqare imager.

Click to expand...

Agreed.

Gary Eickmeier said:

Or are you comparing the square Foveon with the 6mp Canon?

Click to expand...

What I said is that the Foveon sensor in Phil's test can be viewed as the square part of a 6MP sensor. If you had a 6MP sensor and took this square image with it, you would be cutting off 2MP on the right and using the square 4MP part.

--Ron ParrFAQ: http://www.cs.duke.edu/~parr/photography/faq.htmlGallery: http://www.pbase.com/parr/

Ron Parr said:

What I said is that the Foveon sensor in Phil's test can be viewed
as the square part of a 6MP sensor. If you had a 6MP sensor and
took this square image with it, you would be cutting off 2MP on the
right and using the square 4MP part.

Click to expand...

Well what I said was that the Sigma's sensor is going to be like this square sensor with the top and bottom shaved off, to make it only 1512 pixels high.

THEREFORE, its resolution will be somewhat less than the image in Phil's test. That is why Phil cautioned us not to compare this test with other imagers yet.

Gary Eickmeier

Gary Eickmeier said:

Well what I said was that the Sigma's sensor is going to be like
this square sensor with the top and bottom shaved off, to make it
only 1512 pixels high.

THEREFORE, its resolution will be somewhat less than the image in
Phil's test. That is why Phil cautioned us not to compare this test
with other imagers yet.

Click to expand...

No disagreement here.

However, if you wanted to do a "fair" comparison of Phil's results with another camera, a 6MP camera would be the right one to use.

--Ron ParrFAQ: http://www.cs.duke.edu/~parr/photography/faq.htmlGallery: http://www.pbase.com/parr/

Ron,

Whilst I agree with pretty well everything you've been saying about the Foveon sensor (and learned a lot about Bayer interpolation), I'm not sure I agree with you on this one.

By entering the market against 6 MP mosaic chips, Foveon have invited a comparison between those chips and their 3.5 MP CMOS. I don't think there's any doubt that a 6 MP Foveon would eat a 6 MP mosaic chip for breakfast, in fact Phil's posted tests show just that. Seeing those images, I think the 3.5 MP Foveon will be more than competitive with the 6 MP mosaic chips, but there are still a few imponderables, like production tolerances and how they will affect colour channel crosstalk. I will say this though, the images Phil obtained show no problems at all in that area and the 2048 pixel high sensor doesn't just resolve 2000 l/ph, it clearly does it easily and I'd expect extinction to be 2200 or even higher.

Ron Parr said:

Member said:

Well what I said was that the Sigma's sensor is going to be like
this square sensor with the top and bottom shaved off, to make it
only 1512 pixels high.

THEREFORE, its resolution will be somewhat less than the image in
Phil's test. That is why Phil cautioned us not to compare this test
with other imagers yet.

Click to expand...

No disagreement here.

However, if you wanted to do a "fair" comparison of Phil's results
with another camera, a 6MP camera would be the right one to use.

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

Click to expand...

--Richard C. South Australia

I'm sorry, Richard. I'm not sure I understand what you're disagreeing with.

--Ron ParrFAQ: http://www.cs.duke.edu/~parr/photography/faq.htmlGallery: http://www.pbase.com/parr/

Only the issue of what constitutes a "fair" comparison - I don't think a 6 MP mosaic has a snowball's chance in hell of competing with a 6 MP Foveon. The fair comparison will be between, for example, a soon-to-be-available 3.5MP SD9 and a soon-to-be-available 6MP D60. I honestly think the 3.5MP Foveon will have a decisive edge, but, time will tell.

Ron Parr said:

I'm sorry, Richard. I'm not sure I understand what you're
disagreeing with.

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

Click to expand...

--Richard C. South Australia

Gary Eickmeier said:

Well what I said was that the Sigma's sensor is going to be like
this square sensor with the top and bottom shaved off, to make it
only 1512 pixels high.

THEREFORE, its resolution will be somewhat less than the image in
Phil's test. That is why Phil cautioned us not to compare this test
with other imagers yet.

Click to expand...

If you take a 2,000 x 2,000 pixel image in photoshop and crop off the top and bottom 250 horizontal pixel rows have you somehow made what's left more blurry ?