Bayer with NO anti-aliasing (Kodak Pro 14n)

Yeah, I know. But I am just curious if George means spatial in the
third dimension, i.e. depth when he says that the SD9 has 10 million
spatial samples.


Roland

 

How irritating. You say exactly what I say and make
it look like you argue with me. Are you just making
fun of us, or what is your problem?


Roland

 

Not really. What is new is that they made a camera. There
are already devices that uses the wavelength dependent
absorbtion, i.e. color sensors. I have read the Foveon patent,
and really it is (like many patents) questionable. The
ideas were already known, but no one dared to invest money
to make it a reality. It was generally thought as not a good
idea. Foveon has shown that it is possible and might even
become a very good idea, when done right. Good work.


Roland

 

Foveon doesn't sample for color, it captures only full colors, that is the
difference.

 

Ask John, he's the one that says even though Foveon images look better they
aren't.

 

You said "only an idiot" would compare the optical resolution of a 1Ds to an
SD9 with different FOVs. That is the only way to compare optical
resolution.

Cropping is not optical zoom. The SD9 crops a picture from a 35mm lens
projection, the 1DS doesn't, that is what generates the FOV difference.

The reviewer, wasn't an idiot.

 

George - where did I mention Bayer? Nope! So - what difference are
we talking about? Any sampling system needs anti alias filters in
order to avoid aliasing.


Roland

 

In message <bqqqle$f87$>,

No, you pathetic liar; I didn't say anything like that.

I said they look sharp with sharp lenses, but the sharpness is not
legitimate detail. The sharpness is due to the fact that the Foveon
sensors are sampling an area much smaller than their jurisdiction, when
in fact, they should be sampling an area about 4x their jurisdiction.
The results are similar to sampling properly at something like 56mp, and
using Nearest Neighbor to downsize to 3.43mp.

Don't paraphrase what you don't understand.
--

 

No, I didn't. I said comparing a different field of view would be only done
by an idiot.

Bart

 

It's even pretty hard to call them independent measurements. The sensor
response overlap so much that the green and blue sensors recieve about the
same amount energy even if you are only shining a green light on them. They
are highly correlated.

 

That's also true of antialiasing filters, at least the ones that work
by presenting a double image one pixel apart to the sensor. The
frequency response of the filter alone is cos(pi*x) where "x" is in
cycles/pixel, so it has a zero at x=0.5 and other ones at 1.5, 2.5 etc.
But it doesn't attenuate information at 1, 2, 3... cycles/pixel.

So the complete system must depend on the AA filter to provide the
sharp cutoff near 0.5 cycles/pixel, while the area of the pixels
(or microlenses) attenuates around 1 cycles/pixel, and the ultimate
attenuation of higher frequencies is provided by the lens which probably
has a roughly Gaussian frequency response.

Dave

 

An idiotic fraud? A fraudulent idiot?

I see: Bart and I are both right.

Dave

 

Doesn't one of the Nikon cameras do just that?

There's also the hexagonal grid, with the same pattern as the early
triad-gun colour TV sets. Exactly 1/3 of all locations are each red,
green, and blue, all evenly spaced. In some sense, this is the ideal
sensor pattern for both B&W and Bayer-type colour cameras, if you ignore
the difficulty in fabricating the sensor (I'll bet it's a lot easier to
lay out CCD gates in a row/column arrangement than a hexagonal grid)
and processing the data to convert to the standard grid.

Dave

 

You are simply wrong. FOV-cropping is not optical magnification, its
cropping. That is why it is called, guess what? The "crop factor." This
is why the Canon 1Ds has such poor optical resolution compared to the SD9,
the SD9 has the roughly same sensor count packed into a smaller sensor area.
The 1Ds has a wider FOV, but dramatically lower optical resolution...
http://www.outbackphoto.com/artofraw/raw_05/essay.html

 

That aliases a tiny little bit in some small areas of approx 1 in 3000
shots, as measured by objective third party pro reviewers.

 

See above:
I never mentioned cropping, you do.
I never mentioned optical magnification, you do.

An obvious lack of reading skills.

Bart

 

In message <bqsmkk$48j$>,

f45? Dirty filter?

*ALL* sharp SD9 images are completely filled with aliasing, in the
area(s) of focus.
--

 

Absolutely, RGB samples come from mutually exclusive populations. To build
an image, you have a sseparate R channel, a G channel, and a B channel, it
doesn't matter if a sensor is co-located unless it is in the same color
channel, which its not. Foveon takes 10.3M spatially discrete samples,
Bayer 6MP-interpolated only has 6M.

 

Unless they sample from mutually exclusive populations.

 

Correct, and it doesn't make a bit of difference if you layer the 3 color
sensors perfectly aligned, or you slide each of them 1 pixel sideways using
a mosiac then color interpolate, they don't sample the same populations so
the photosites in each channel are just as discrete either way.