I'll bet it equals the IQ of a 10 MP Bayer rather than an 8 MP.

I don't care to re-hash the reality of the clear lack of spatial resolution
in the foveon...and the "invented" lines inherent in its lack of a sensor
filter...  Sigma renderings of a Large number of textures is utter crap.  No
thanks.

However, try this: Take any image in photoshop, open it three times; in
one copy, add noise only to the blue channel, in the second only to the
blue, and in the third, only to the green (same noise distribution).
Compare the results. Are all channels contributing equally to resolution
(as perceived by the eye/brain combination)? What do we conclude from this?

Not that it's not better to sample 3 colours from each of N sites than 1
colour from each of N, just shows that 3 colours from each of N/3 is not
the same as 1 colour from each of N (in terms of perceived resolution;
if you have a sharp transition between red and blue, the Bayer CFA will
get give artifacts; but of what spatial extend? Things are not so simple).

Well regarded by who - Homer Simpson?

I heard that there is software that is able to shift colors <g>.

So image/camera quality can't be measured by comparing a single pixel.

-Michael

You won't see any pros using these things, just like only a small
handful used their predecessors. And like the Sigmas before it, the
dent in the marketplace will be insignificant.

I trust those who have seen the results.  I suspect that no test results or
samples or whatever will change minds like the OP.

The OP accuses some of being a Sigma fanatic, she can be accused of being an
anti-Sigma fanatic just as easily.

When the camera performs as well as a 10-12 MP camera, where is the insane
and outright absurd rationalization?

It's obsessing and it's evident by the aburd statements that are made
without having any real knowledge or evidence to back them.

Randall, you know you weren't the person I was responding to when I posted
this.

I won't quote back any marketing spiel.  I have an SD9 and a D50.  I can
tell you without a doubt that the SD9 outresolves the D50.  We're dealing
with two different types of sensors, judging one against another in terms of
MPs is not the way to do it.  Results are far more indicative than anything
else for me and the results I have seen are enough for me.

The Sigma 50mm EX and 105mm EX are two of the sharpest lenses you can get.
Perhaps you've been a victim of spotty QA.

Results, results.  I see how the 3.4 MP SD9 is better then the 6MP D50, I've
heard from users who shot with the SD14 and their reports indicate that it
is in the league of any 10MP Bayer.

In the right conditions the SD10 can produce very clean high ISO shots but
it definitely not it's stron-suit by any margin.  From those who've seen the
24x36 SD14 prints, which were produced by some of us lowly Sigma users - not
pros - the improvements in handling noise are big.

I think that someone who has, in your words, "contempt" for an inanimate
object like a camera is a bit obsessive about it.

I am not sure if there is much point in participating in these type of
discussions. The opnions on both sides seems to well entrenched, so it
is likely that what I am about to say will make a difference.  Also,
as someone whose opinion is based upon actual use of a Foveon camera,
as well as several Bayer cameras, (see sig) - I will probably be
written off as brainwashed anyway.

But against better judgement, here is my article ...

First since I have not used the SD14, I will not comment on the 10-12
Mpx (equivalence claim), but address the more general problem of
comparing Mpx number between Foveon and Bayer sensors.

I own a Sigma SD10.  It is a camera with 3.4 Mpx a Foveon sensor.  My
subjective findings, based upon numerous side-by-side comparions with
images shot with a 6.3 Mpx Canon EOS 10D camera is that the 3.4 Mpx
Foveon resolves on par (or better) than the 6.3 Mpx Bayer sensor in
the Canon.  Both were tested with the same lens, a 50mm f/1.8 Nikkor
stopped down to f/8, (used via adapter, and focused manually).

How can that be?  3.4 Mpx < 6.3 Mpx, right?

However, a 6.3 Mpx bayer doesn't resolve 6.3 Mpx.  For starters: the
particular Bayer design mandates that it has an AA filter.  Otherwise
false colour moire becomes a huge problem.  Depending on how aggresive
the AA filter is, spatial resolution is reduced.  A filter that
removes all frequencies above the Nyquist limit would half the spatial
resolution, but "real" filters are not a as aggressive as that, so the
effect of putting an AA-filter in front of a 6.3 Mpx bayer will
probably reduce it to something bewtween 3.5 and 4.5 Mpx
resolutionwise.

The Foveon sensor design doesn't create false colour moire, so there
is no AA-filter in my SD10.  That means that is resolution of a Foveon
sensor is the number of megapixels that follows from the number of
spatial locations where it captures light (3.4 million in the case of
the SD10).

As a number of well respected contributors without doubt will tell me,
Foveon's design isn't immune to other aliasing artifacts, such as
regular (monochrome) moire, and staircasing - so it really /should/
have an AA-filter.  Yes and no.  It is true that with certain
subjects, aliasing is a problem.  However, for a number of subjects,
it isn't a problem, and that is the type of subjects I use it to
photograph.

However, the AA-filter is not the whole story.  I have also compared
a SD10 (3.4 Mpx) to a Kodak DCS 460c studio camera (6.0 Mpx).  The
Kodak is a fairly rare animal, in the sense that it is a camera
with a Bayer sensor /without/ and AA-filter.  I.e. both cameras are
without an AA-filter, and still the Sigma produces sharper images.

To understand this, we need to look closer at how the Bayer image
is created. A 6.0 Mpx Bayer sensor has 3.0 Mpx green photosites, 1.5
Mpx blue, and 1.5 Mpx red.  It measures luminance in all 6 million
photosites, but since colour information isn't available at each
photosite, it need to interpolate the missing colour information to
create a 6.0 Mpx image.

People that argue that 3.4 Mpx Foveon < 6.0 Mpx Bayer tend to forget
that a lot of the colour information presented in the 6.0 Mpx Bayer
(66 % of it, mathematically speaking) comes from interpolated data.

I haste to add that speaking mathematically does not reveal the whole
story. The human visual system is much more sensitive to variation in
luminance compared to chroma, and for this reason, Bayer interpolation
works extremely well and much better than the math predicts.  However,
some of the information in the "Mpx" quoted in the marketing
literature by Bayer manufacturers /are/ interpolated.  So there is
some hypocracy when people attack Sigma for inflating their
"marketing" Mpx count by referring to interpolated figures, but not
mentioning that the numbers quoted by Canon, Nikon et al, also do this
(to some degree - not necessarely the same extreme degree as Sigma's
figures).

To return to your original question: How is it physically possible for
a 4.6 Mpx Sigma SD14 resolve on par with a 10-12 Mpx Bayer camera?

First - I am not saying it does (I am waiting for the tests), but the
claim is not as ludicrous as you think it is.  You see: The 10-12 Mpx
Bayer doesn't really resolve 10-12 Mpx.  The presence of the AA-filter
probably cuts it effective resolution to 5.5-8 Mpx, and the colour
interpolation that is part of Bayer demosaicing probably knocks
another Mpx of its effective resolution.  So a 10-12 Mpx Bayer shall
probably resolve something like 4.5-7 Mpx non-interpolated pixels - so
this is the figure you should compare to the non-interpolated 4.6 Mpx
produced by the SD14's Foveon sensor.

For those that wants hard data, here is a paper that compares
the effective resolution of Foveon vs. Bayer technologies:

  http://www.foveon.com/files/ResolutionforColorPhotography.pdf

For those that don't want to read the whole thing, the most
interesting data is on page 4 (figure 3 og 4), and page 6 (figure 5
and 6). Both compare the measured resolution of a Sigma SD10 (3.4 MPx)
and an unnamed 6.3 Mpx Bayer camera (however, the number of pixels
matches the sensor used in Canon EOS 10D/300D).

For the record: One of the two authors work for Foveon (the other
works for the German research foundation Stiftung Warentest), but the
paper documents the way the measurements are done, so the whole thing
is open to peer review and independent replication and verification/
falsification.

Here you have an intractable problem. I have Minolta's excellent 100mm
f2.8 macro, and like the Nikon 105mm, and like the Sigma, it does
exactly the same - stop down below f8, even at 1:2-1:1 macro distances,
and the diffraction begins to take the edge off resolution. Take it down
to f22 or f32, which I have to do until I replace the controller for my
ringflash (only usable on full power right now) and the pix are not
usable. Digital sensors show me what film might not reveal, the critical
sharpness present at f5.6 (about the best for most of these lenses) is
destroyed. So it's a choice, either get depth of field, or ultimate
sharpness in a narrow plane.

It does not run out of res, it hits a physical factor which will always
be present now we are using the discrete pixel well size associated with
digital sensors, instead of fuzzy gelatin analog film. Most expert macro
photographers are finding the same thing regardless of the brand or type
of macro.

I just acquired a 25mm f2.5 micro bellows head lens after 20 years
without one - and guess what - it's diffraction limited in a serious way
for digital shooting.

As for your dismissal of the Foveon sensor and the Sigma, it's hard to
understand why you should do so. The native (true 3.4 megapixel) image
from the original Foveon has remained, for me, a benchmark against which
to judge Bayer sensor results and the quality of raw convolution
algorithms for deBayering. Only because it doesn't need them! It is as
pure an image as the three-line RGB from our early Leaf scanning back
used to be, or three-pass from Dicomed and other pioneers.

There are issues present with aliasing of diagonal structures and
rendering some straight edge contrasts (one of my first test shots with
the SD9, of Melrose Abbey, apparently doubled the number of steps in the
lead roofing pattern). Checking the 24 x 36 prints at photokina from the
SD14, these are not solved - they are still present, distinctive and
unwanted stepping of certain edges and textures. Hair can really show
this up.

However, once you get away from man-made structures and into the more
random and fractal world of trees, leaves, grass, skin, rocks, etc the
regular geometry of the Foveon captures a better impression of textures
and forms than typical Bayer sensors, and the result stands enlargement
to the kind of degree they claim.

Personally I will support and encourage Sigma and Foveon alongside the
other cameras we use. For one thing, it is the only technology which
does NOT show the diffraction sharpness loss problem associated with
extremely small pixel wells, and one chance we have (if the design can
be progressed) of full frame sensors with relatively big pixels.

David

Helen skrev:

This is because you haven't understood to what extent anti-aliasing and
Bayer interpolation, neither present in a Foveon sensor, decrease
actual resolution in a conventional sensor. With a sensor of the Foveon
type, you get full resolution value for your megapixels. With a Bayer
sensor, you don't - it's inherent in a construction that only records a
single channel at a given pixel site. There are several other
advantages to a Bayer type sensor, but information efficiency in terms
of resolution/pixel number isn't one of them.

There are certainly others. I have a friend who sold his Canon 14mm 2.8
L and bought Sigma's corresponding lens, and who claims that the Sigma
is not only equal, but actually sharper. At like half the price.

Sigma generally seems to have a good hand with fixfocal wides. I have
tried their 20mm 1.8 on a couple of occasions. It's a surprisingly good
lens for such a reasonable price.

I own this lens, and I've never noticed any sharp transition of
resolution at f/13. I shoot routinely at f/22, and, while diffraction
necessarily kills some fine resolution at that aperture, it's not more
than what could be expected from any lens.

Maybe you just got a bad copy?

Jan Böhme

I know the background to this, at least in part, and the Foveon was
offered to and considered by two major names. One of them was supposed
to run with it. However, that company relied on Sony for consumer
digital camera sensors, and Sony put great commercial pressure on them
to abandon the Foveon chip, and use their 'new' 6 megapixel CCD.
Accordingly, they did so, delaying their camera by 2 years in the
process, so that when it appeared it was based on a surprisingly old
body design.

Foveon had to find a camera maker partner - other than Canon, who had no
 interest - which did not make a line of compact digitals, or have an
existing vital relationship with Sony which would have been placed under
stress by choosing a National Panasonic backed technology.

Sigma had no compacts. They had no other DSLRs. They bought sensors from
no-one, but made digital SLRs for everyone. They were already making
DSLRs under various names (as a customiser, using parts of the maker's
gear and adding new components of their own). They had all the
experience and none of the disadvantages of being commercially tied to
an implacably hostile Foveon competitor -

so Sigma got the deal. This is also why Polaroid ended up using Foveon -
no history of dependence on other technology.

Two years after Foveon first planned to be there.

David