Digital BW, The Print

In a message dated 12/29/2003 5:10:22 PM Pacific Standard Time, 
DigitalBlackandWhiteThePrint@yahoogroups.com writes:
It would not be a problem with three
separate CCDs capturing every primary color for every pixel, or the
equivalent (like a Foveon, if it were ever perfected).


I have been using the original 3-chip Foveon (1999) for 4.5 years now. I have 
also experienced the SD-9 with the X3 chip, which I suggested they develop.

What aspect are you referring to that needs perfecting? It outperfoms all 6 
megapixel monochips right now in all respects except for ISO higher than 400. 
In fact, it gives the 11 megapixel Canon a run for it's money.

I can't think of a better camera for Black and White work for less than 
$1,000 AND it has the best color response out there. It's unbeatable for Macro Work 
and Sigma lenses are very good and 1/3 the price of Nikon/Canon.

Claude Jodoin


[Non-text portions of this message have been removed]

> From: claudej1@... [mailto:claudej1@...]
>
> What aspect are you referring to that needs perfecting? It
> outperfoms all 6
> megapixel monochips right now in all respects except for ISO
> higher than 400.
> In fact, it gives the 11 megapixel Canon a run for it's money.

Judging by the few SD-9 samples I've downloaded, I'd prefer an antialiasing
filter over it.

--

Ciao,               Paul D. DeRocco
Paul                mailto:pderocco@...

You need to differentiate preference over technical performance. From 
the reviews I saw the SD9 had better resolution an any of the current 
crop of 6 meg DSLR's and if you stay within the dynamic range of the 
sensor the color resolution seems much better. Whether you like it or 
not is a different issue. I have a friend who still perfers the sound of 
vinyl LP's over CD's.

Truman

Paul D. DeRocco wrote:

> > From: claudej1@... [mailto:claudej1@...]
> >
> > What aspect are you referring to that needs perfecting? It
> > outperfoms all 6
> > megapixel monochips right now in all respects except for ISO
> > higher than 400.
> > In fact, it gives the 11 megapixel Canon a run for it's money.
>
> Judging by the few SD-9 samples I've downloaded, I'd prefer an 
> antialiasing
> filter over it.
>
> --
>
> Ciao,               Paul D. DeRocco
> Paul                mailto:pderocco@...
>

> From: Truman Prevatt [mailto:tprevatt@...]
>
> You need to differentiate preference over technical performance. From
> the reviews I saw the SD9 had better resolution an any of the current
> crop of 6 meg DSLR's and if you stay within the dynamic range of the
> sensor the color resolution seems much better. Whether you like it or
> not is a different issue. I have a friend who still perfers the sound of
> vinyl LP's over CD's.

Well, sure. It's an unavoidable trade-off: you can either have the extra
sharpness, or you can smooth out the jaggies. Since the more popular
digicams have a filter, people can mistakenly assume that the Foveon chip is
somehow unusually sharp, but it really isn't. Don't get me wrong, I think
it's a fabulous technology, and it could well be the wave of the future, but
I prefer having a filter of the sensor, because it avoids some distinctly
digital looking artifacts.

--

Ciao,               Paul D. DeRocco
Paul                mailto:pderocco@...

claudej1@... writes:

> What aspect are you referring to that needs perfecting?

Many sources say that it still needs work.  I haven't tried it myself.
Obviously, the fundamental idea of capturing all three primary colors at
every photosite is indisputably superior to using a matrix color filter.

There has been some modifications to the sensor that goes into the SD10. 
Maybe some of the issues have been addressed. I guess we will soon see. 
The color matrix filter over an intensity CCD is basically three (or 4) 
offset sensors (one for each color) each with 1/3 (or 1/4) the the 
native resolution. the final image is produced by interpolation of these 
slightly offset images. The final  image cannot contain the information 
to produce the resolution or fidelity of the native resolution of the 
sensor. If the information is not there it can not be produced by 
interpolation no matter how sophisticated the interpolation algorithm. 
Interpolation is a smoothing process operating on the information that 
is there and cannot produce additional information. So it is somewhat 
misleading to advertise the resolution of these cameras at the native 
resolution of the sensor. It is just as valid for foveon to advertise 
their sensor as 3 X or about 10 megs as for Canon or Nikon to advertise 
theirs at the native resolution. There are a lot of shell games going on 
in the advertising of digital cameras.

Truman

Anthony G. Atkielski wrote:

> claudej1@... writes:
>
> > What aspect are you referring to that needs perfecting?
>
> Many sources say that it still needs work.  I haven't tried it myself.
> Obviously, the fundamental idea of capturing all three primary colors at
> every photosite is indisputably superior to using a matrix color filter.
>
>
>

> From: Truman Prevatt [mailto:tprevatt@...]
>
> There has been some modifications to the sensor that goes into the SD10.
> Maybe some of the issues have been addressed. I guess we will soon see.
> The color matrix filter over an intensity CCD is basically three (or 4)
> offset sensors (one for each color) each with 1/3 (or 1/4) the the
> native resolution. the final image is produced by interpolation of these
> slightly offset images. The final  image cannot contain the information
> to produce the resolution or fidelity of the native resolution of the
> sensor. If the information is not there it can not be produced by
> interpolation no matter how sophisticated the interpolation algorithm.
> Interpolation is a smoothing process operating on the information that
> is there and cannot produce additional information. So it is somewhat
> misleading to advertise the resolution of these cameras at the native
> resolution of the sensor. It is just as valid for foveon to advertise
> their sensor as 3 X or about 10 megs as for Canon or Nikon to advertise
> theirs at the native resolution. There are a lot of shell games going on
> in the advertising of digital cameras.

Have you ever actually examined, at the pixel level, the results of the
Bayer interpolation performed by a camera like a Canon 10D? You can talk all
the theory you want, but they've developed some stunningly smart algorithms
for edge detection, etc. There's nothing misleading about calling a 6
million site sensor with a Bayer pattern a 6 megapixel sensor. The
algorithms really do work, on real-life images. It would be far more
misleading for Foveon to say that their 3 megapixel sensor has three times
the resolution as the number would suggest, and I don't believe they make
such a claim.

--

Ciao,               Paul D. DeRocco
Paul                mailto:pderocco@...

No matter what your eye tells you, information cannot be created.  Given 
a fairly complex image, the interpolation looks okay, but take it to the 
bench for testing and it will show at that point. Give it a particularly 
difficult image and it will have some artifacts (it is a not only a 
smoothing operation it is non-linear and non-linear operations will 
product artifacts). 

As far as resolution comparision check 
http://www.dpreview.com/reviews/sigmasd9/page23.asp where the reviewer 
pegs the SD9 better than the six meg sensors out there.  Its resolution 
is better according to the table in the above review to the 6's out 
there. I've seen ads for the SD9 that claims an "effective" 10 meg 
resolution. One was in the last B&H catalog I received.

Truman

Paul D. DeRocco wrote:

>
>
> Have you ever actually examined, at the pixel level, the results of the
> Bayer interpolation performed by a camera like a Canon 10D? You can 
> talk all
> the theory you want, but they've developed some stunningly smart 
> algorithms
> for edge detection, etc. There's nothing misleading about calling a 6
> million site sensor with a Bayer pattern a 6 megapixel sensor. The
> algorithms really do work, on real-life images. It would be far more
> misleading for Foveon to say that their 3 megapixel sensor has three times
> the resolution as the number would suggest, and I don't believe they make
> such a claim.
>
> --
>
> Ciao,               Paul D. DeRocco
> Paul                mailto:pderocco@...
>
>

> From: Truman Prevatt [mailto:tprevatt@...]
>
> No matter what your eye tells you, information cannot be created.  Given
> a fairly complex image, the interpolation looks okay, but take it to the
> bench for testing and it will show at that point. Give it a particularly
> difficult image and it will have some artifacts (it is a not only a
> smoothing operation it is non-linear and non-linear operations will
> product artifacts).

But the eye is all that matters. Your statement makes me wonder whether
you're using your camera for art or for lab measurements.

I know what constitutes a "difficult" image, because my old Minolta D7
didn't have such a good interpolator. It would show color noise on fine
lines, like individual hairs. The 10D's internal interpolator, and the one
in the latest Adobe Camera Raw, are much better than that. They can indeed
render detail down to the pixel level, so they're not simply smoothing, but
I have yet to see anything out of them that the eye would recognize as a
wrong pixel. Now, if you took a picture of some sand, roughly matching the
grain size to the pixel size, I'm sure it wouldn't render the color of each
grain as precisely as the Foveon chip, but who cares, as long as it looks
like sand? Where it matters is where the eye would see the difference.

> As far as resolution comparision check
> http://www.dpreview.com/reviews/sigmasd9/page23.asp where the reviewer
> pegs the SD9 better than the six meg sensors out there.  Its resolution
> is better according to the table in the above review to the 6's out
> there. I've seen ads for the SD9 that claims an "effective" 10 meg
> resolution. One was in the last B&H catalog I received.

Yes, that's BECAUSE there's no anti-alias filter over it.

--

Ciao,               Paul D. DeRocco
Paul                mailto:pderocco@...

Paul D. DeRocco wrote:

>
> But the eye is all that matters. Your statement makes me wonder whether
> you're using your camera for art or for lab measurements.
>
> I know what constitutes a "difficult" image, because my old Minolta D7
> didn't have such a good interpolator. It would show color noise on fine
> lines, like individual hairs. The 10D's internal interpolator, and the one
> in the latest Adobe Camera Raw, are much better than that. They can indeed
> render detail down to the pixel level, so they're not simply 
> smoothing, but
> I have yet to see anything out of them that the eye would recognize as a
> wrong pixel. Now, if you took a picture of some sand, roughly matching the
> grain size to the pixel size, I'm sure it wouldn't render the color of 
> each
> grain as precisely as the Foveon chip, but who cares, as long as it looks
> like sand? Where it matters is where the eye would see the difference.


I want the best possible sensor (be it film or a CCD) before I start. 
This is why I normally use a 4x5 or if I use a 35 mm I use a fine grain 
thin emulsion film.  I don't shoot low light, I don't shoot sports. For 
the subjects I tend to shoot, quality image is essential. That way I am 
not too concerned about being limited by the the sensor. My stero system 
has response up to 20 kHz. I can't hear 20 kHz, but I can hear the 
difference if I filtered the output to 10 kHz before sending it to the 
speakers on some music. Same thing with image "frequency."

>
>
>
> Yes, that's BECAUSE there's no anti-alias filter over it.

Yep and the reviewer commented that it didn't seem to require one.

Truman

Paul D. DeRocco writes:

> Have you ever actually examined, at the pixel level, the results of
> the Bayer interpolation performed by a camera like a Canon 10D?

Yes.  The colors are smeared.

> You can talk all the theory you want, but they've developed some
> stunningly smart algorithms for edge detection, etc.

No amount of smartness in algorithms can abridge the laws of information
theory.

> There's nothing misleading about calling a 6 million site
> sensor with a Bayer pattern a 6 megapixel sensor.

That depends on what you mean by six megapixels.  Six megapixels from a
Bayer filter is very different from six megapixels from scanned film, or
six megapixels from a digicam that captures all colors at every pixel
(like the Foveon).

> The algorithms really do work, on real-life images.

They do not and cannot recover information that was missing from the raw
capture.

> It would be far more misleading for Foveon to say that their
> 3 megapixel sensor has three times the resolution as the number
> would suggest ...

I don't see how.

Truman Prevatt writes:

> As far as resolution comparision check
> http://www.dpreview.com/reviews/sigmasd9/page23.asp where the reviewer
> pegs the SD9 better than the six meg sensors out there.  Its resolution
> is better according to the table in the above review to the 6's out 
> there. I've seen ads for the SD9 that claims an "effective" 10 meg 
> resolution. One was in the last B&H catalog I received.

Sliding back towards the topic of the mailing list:  A Foveon should
provide far better black and white results than a typical sensor with a
matrix filter.

Paul D. DeRocco writes:

> But the eye is all that matters.

Viewing conditions matter just as much, and what satisfies the eye under
one set of conditions may be hopelessly inadequate under other
conditions.

> Your statement makes me wonder whether
> you're using your camera for art or for lab measurements.

Your close approach to a personal attack makes me wonder how much
objective data and reasoning you can muster to support your position.

You're entitled to an opinion, but there's no shame in admitting that it
_is_ just an opinion.  And subjective opinions cannot be objectively
proved "right" or "wrong."

> They can indeed render detail down to the pixel level ...

No, they cannot.  They can pretend, but they cannot show detail that
wasn't in the original capture.

> ... I have yet to see anything out of them that the eye would
> recognize as a wrong pixel.

Compare the interpolated image to the original scene, and you may be
surprised.

> Now, if you took a picture of some sand, roughly matching the grain
> size to the pixel size, I'm sure it wouldn't render the color of each
> grain as precisely as the Foveon chip, but who cares, as long as it
> looks like sand? Where it matters is where the eye would see the
> difference.

The eye will see the difference if it looks closely.  Then it matters.

It's interesting to see all the effort being wasted on trying to prove
that matrix filters are just as good as systems that capture every color
for every pixel.  The fact is, they aren't.  And the same people who are
trying to prove that they are will probably be the first to say how
terribly inferior matrix filters are once they have a camera in hand
that really does capture all colors for all pixels.

I've seen this all happen in video, years ago.

Truman Prevatt writes:

> Yep and the reviewer commented that it didn't seem to require one.

Aliasing is far less of a problem when you have all three colors for
every pixel.  A Bayer filter will show aliasing of blue and red detail
even at frequencies four times lower than the pixel frequency.

For those of us who prefers actual real life samples to long winded
arguments these pages are pretty interesting:

http://www.dpreview.com/reviews/sigmasd9/page20.asp
http://www.dpreview.com/reviews/sigmasd9/page22.asp

It's comparisons between the SD9 and D60 where the SD9 images were
enlarged to 6 megapixels. They seem to prove Anthony's points very well.
I was surprised to see that the SD9 images could actually hold sligthly
more detail and were sharper than the D60 images.

I also this is very relevant to BW. Especially when you consider channel
mixing techniques where you're often taking most of the data from a
single channel to achieve the final result.

--
Daniel Staver
http://daniel.staver.no

Daniel Staver writes:

> For those of us who prefers actual real life samples to long winded
> arguments these pages are pretty interesting:
>
> http://www.dpreview.com/reviews/sigmasd9/page20.asp
> http://www.dpreview.com/reviews/sigmasd9/page22.asp

The SD9 looks a lot better to me.  I see what looks like chromatic
aberration (and the review agrees), but that is probably elsewhere than
on the sensor (the D60 just is too blurry to show it).

--- In DigitalBlackandWhiteThePrint@yahoogroups.com, claudej1@a... 
wrote:
> I have been using the original 3-chip Foveon (1999) for 4.5 years 
now. I have 
> also experienced the SD-9 with the X3 chip, which I suggested they 
develop.

Which chip are you talking about? The prototype? ANd how are you 
using it, with what camera? Was it the square shaped 2000x2000 pixel 
thing?

> What aspect are you referring to that needs perfecting? It 
outperfoms all 6 
> megapixel monochips right now in all respects except for ISO higher 
than 400. 

That was not my experience: I had the SD9 here for a week- end, and 
while there is a lot of detail resolution, the noise was visible from 
ISO 100, getting a problem in the darker shadow areas from ISO 200 
and too much for my taste at ISO 400. Due to the higher noise the 
photos didn't perform too well in up- rezzing either.

> In fact, it gives the 11 megapixel Canon a run for it's money.

A nice story, but I yet need to see samples that can convince me. I 
like such David versus Goliath stuff. Can you direct me to samples 
that 
illustrate your claim?

> I can't think of a better camera for Black and White work for less 
than 
> $1,000 

I use the Fuji S2 and have had very good results up to 13x19 
converted to black and white. The few examples of the SD9 that I 
uprezzed didn' t look as good as the S2´, although its detail 
resolution in the original output size (3,3MP) was quite amazing. 

> and Sigma lenses are very good and 1/3 the price of Nikon/Canon.

This is a general claim that doesn't make sense, as it is a well 
known fact that for one manufacturer there are always better and 
worse lenses, for example the relatively new 24-135mm Sigma only got 
mediocre ratings. Even the higly praised 70-200 has often been 
reported to be a bit too soft.

Regards and a satisfying and happy new year,

Bernhard

This thread has wandered off the topic for this group, digital B&W printing.
Please end off or take the discussion off-list.

Thank you for your cooperation.

Group Moderators



"Users who repeatedly make off topic posts may have their ability to post
restricted or their membership terminated at the discretion of the group
Owner and/or Moderators. By participating in this forum users agree that
what constitutes an off topic post is solely at the discretion of the group
Owner and/or Moderators This may sound harsh but from experience it is
essential to maintaining a high quality and useful forum."