Steve's Digicam's classifies the SD-9 and SD-10 as 10.2MP

Message finally received. The first domino has fallen.

"The improved 10.2 megapixel (approx. 3.43 million each red + green + blue
pixels) image sensor doubles the sensitivity and maximum exposure times and
offers increased dynamic range over its predecessor. "
http://www.steves-digicams.com/2003_reviews/sigma_sd10.html

Don't know where the ".2" comes from with 10.29M RGB sensors, but the SD-9
and 10 are finally properly classified.

Good for Steve's Digicams.

 

Looks like a boo-boo. According to the review, the SD10 can produce
images up to 2268x1512 pixels; when I do the math, that's a wee bit short
of 10,200,000.

 

Sorry, George.....The Sigma/Foveon cameras are NOT10Mp according to
the prevailing standard. I can back this up with all the theory and
data that anyone cares to listen to.

In common practice, xx Mpixels refers to the number of independent
spatial samples that can be used to create spatial resolution---ie
detail. Sigma/Foveon simply are not following this standard.

We desperately need some kind of official standards (like maybe ISO
film speed). Until we get them, we owe it to ourselves to follow the
de-facto standards---as maybe 80-90 of the vendors are doing.

Shame on Steve's or dpreview or anyone else for perpetuating and
credibleizing (there's a good one!!) the claims of a small minority
trying to find their niche.

-mark
*****************************************
digital photos, more and better computers,
and never enough time to do the projects.
Private e-mail: Just say no to No

 

Word has it that on 28 Oct 2003 02:18:57 -0800, in this august forum,
(Leicaddict) said:

Troll, are you actually stupid enough to think that your lame
'forgeries' of us will trick Leicaaddict & I into fighting? If so,
you're even dumber than I've been thinking, (if that's possible).

Kisses,

Lionel.

 

Not a good first attempt.

 

The SD-9/10 takes 10.3M indepentent spatial samples, so sure they are.
Here's the chart again...
http://www.pbase.com/image/22273598

Here is how it works, for real. To sense 1 full color Bayer uses a 4 sensor
set...
-----
RG
GB
-----
It outputs 4 interpolated color pixels which all center on that 1 sensed
color. Any color variation between those 4 pixels is artifactual (a much
better word than credibleizing). The extra G is wasted, because in a 2x2
grid one of the 3 primaries has to be wasted. So it uses the extra G to
tune overall luminance a little, then throws the extra-green info away.

Foveon uses 3 sensor to sense 1 full color...
-----
X
-----
Where X is a complete RGB set aligned in depth. It outputs 1 pixel with 1
non-interpolated color (with a known, not sampled, luminance value).

The end result is the Bayer scales all of its images up by 4X, Foevon
doesn't. Simple scaling up (interpolating) doesn't change optical
resolution one bit. So theoretically, the the SD-9 is exactly the same,
optically, as a 13.7MP Bayer, which not coincidentally is the SD-9's
(optional) interpolated MP output.

But... only when color is sensed. For B&W targets the Bayer performs much
better, color interpolation is closer to 100% effective when no color is
involved.

 

Word has it that on Wed, 29 Oct 2003 07:30:48 +0900, in this august

Sorry, but this is utter bullshit. You clearly don't understand the
meaning of the phrase: "independant spatial samples". Three photosites
stacked on top of each other only counts as a single spatial sample.

 

I would assume that any of the demosaicing algorithms used in current
cameras are far more sophisticated than your simple model...all you
have to do is read a few of the papers, and there are hundreds to see
the options. None of the ones I have read discard any G data and use
moving sampling structures of (many) adjacent pixels with weighting
schemes. No one knows what any of the major manufacturers have
incorporated in their firmware but your illustration is clearly bogus.

 

SNIP

sensor

Wrong!

Depending on the actual algorithm, *each* pixel uses its 8 independent
surrounding pixels (or more) to reconstruct the RGB triplet.

Bart

 

Nope--sorry. Independent spatial samples means they are taken in different
parts of the scene. Basic sampling theory--well documented. contact me
privately for the references

 

It proves 3 things.

--Sigma and Foveon are using marketese to "boost" their sensor
--Steve's D's have bouhght the line, or been paid to do so
--Steve's D's are just as gullible as you are.

 

It does not throw the extra green away...

George: we know how to count. That's why we can count 3.4 Mpix on the
Foveon and 11 Mpix on the 1Ds. 'Cause there ARE 11 Mpix on the 1Ds.

 

Interesting response.

 

Same amount of raw optical data, any way you juggle it.

 

The Foveon has better spatial separation between color samples, much better,
almost 2X better.

 

Count the pixels in this SD-9 image...

http://www.pbase.com/image/22500119/original

What do you get?

 

There is another possiblitly.

--You didn't know Bayer counts 1/3rd color sensors as full pixels when
rating their cameras.

 

Word has it that on Wed, 29 Oct 2003 11:08:33 +0900, in this august

Gotta love the way you snipped out the *reasons* why your claim is
obvious bullshit. You obviously don't have answers for them:

 

Whatever the reply George, the SD-9 is sampling at 3.4 million PLACES.
Period. The depth of information may be different, but that is all.

 

X3