This discussion thread is all over the place, with all kinds mumbo
jumbo speculations about how to compare Foveon vs mosaic sensors.
I'm posting in this part of the thread because it has had the most
recent activity.
I think that technically any 6mp camera will have the edge over
even a 3mp Foveon in absolute ability to distinguish that fine
details exist. I think the Foveon has will have a big advantage in
the quality of any detail it shows. To try to assign a useful
ratio is fruitless, because you're comparing carrots to pineapples.
All of the issues of Foveon vs Bayer Mosaic sensors relate to human
perception of luminance and color, plus issues of artifact, noise,
aliasing, interpolation, etc.
First, related to the discussion of RGB patterns, there are all
kinds of good sample pictures of mosaic sensors at this web-site.
See
http://www.dpreview.com/news/0202/02021101foveonx3.asp
or the DPReview Glossary
http://www.dpreview.com/learn/Glossary/Camera_System/Colour_Filter_Array_01.htm
If you compare a 3mp Foveon to a 3mp mosaic, the mosaic typically
has 1/2 the green, and 1/4 the red and blue pixels. There tends to
be a confusion in this whole thread about linear resolution vs
mega-pixels. If you double the horiz & vert resolution, you will
end up with 4 times as many pixels. That is: a 12mp camera has
double the resolution of a 3mp camera.
Technically, if you photograph a B&W high-rez image (with high
frequency detail) in good well balanced light (probably day-light)
using the B&W mode of the color mosaic camera, you should be able
to approach the theoretical resolution of the camera. Why? Because
the white will be equally sensed by any red, green, or blue sensor.
Similarly, black will be sensed equally by any sensor. In B&W
mode, the camera's software "assumes" that rapid changes between
pixels are due to luminance, and just outright eliminates any
color. However, in color mode, the camera has absolutely no way to
distinguish whether rapid changes from the red to the green pixel
are due to color, or due to high frequency luminance detail. So
the high resolution is present, but tends to fall apart in color
mode. This gets even more murky when the high resolution detail is
very colorful (flowers, birds, ...) So what? A Foveon sensor
should theoretically give the same resolution in B&W or color mode,
and will give you it's true and full resolution under ALL
circumstances.
I happen to own a 3-CCD video camera, in addtion to a Canon G1.
3-CCD cameras use very different technology to achieve a similar
result to the Foveon chip. The 3-CCD cameras use dichroic prisms
to split the red, green, and blue light to three B&W CCDs
(non-mosaic). Although my video camera gives only 640x480
resolution still images, each pixel has red, green & blue, and the
quality of the images is excellent, other than the low-rez
limitation. The 3-CCD cameras are very good at what they do, but
cannot easily reproduce certain colors, for example a deep violet
iris shows up as a deep blue. The red cones in the human eye have
a secondary sensitivity peak in the violet end of the spectrum.
The 3-CCD cameras do such a good job of separating the red from the
blue ends of the spectrum that the very deep violet cannot be
properly represented. However, most purples actually reflect both
red and blue ends of the spectrum, and look fine with a 3-CCD
camera. I will be very interested to see how the Foveon sensor
deals with certain extreme colors.
I personally believe that MUCH of the "noise", and certainly all of
the color moire patterns in mosaic CCDs are related to the
conversion from adjacent RG&B pixels into full resolution images,
without proper anti-aliasing filters. Some of the high end cameras
do actually have mechanical anti-aliasing (blurring) filters in
front of the CCD. Classical engineering training teaches that you
MUST ELIMINATE any frequencies above 1/2 the sampling rate (Nyquist
criterion) BEFORE SAMPLING the signal. Otherwise, you WILL GET
aliasing (moire patterns the photo domain). Many moire patterns
are impossible to correct for. The unfortunate side effect of a
properly designed anti-aliasing filter is that it WILL blur your
picture, but only in a way that is similar to the human eye or
film. For example, using your eyes, if you walk away from a subject
with a striped shirt, as you approach a point the lines start to
lose contrast, then start to blur together. As you continue to
move away, the shirt will simply look like a solid color. Not so
with most digital cameras. Even a Foveon sensor has the
possibility of exhibiting aliasing, but not the severe chromatic
aliasing very common with digital cameras.
People can go on and on about 2x or 3x or resolving power or 56% or
75% or sharpness or file size or compression. The fact of the
matter is that this group has grown accustomed to the type of
artifacts inherent in the mosaic CCDs. I predict that many readers
here will be pleasantly surprised by the clarity and cleaness of
Foveon type images, even though they are lower in resolution. I
don't know how the industry will play out. It wouldn't surprise me
if the industry giants (camera or CCD manufacturers) try to quash
or circumvent Foveon, but I applaud Foveon's ingenuity and efforts.
One more little thing about RAW in a Foveon vs mosaic camera. A 6mp
mosaic has to store 6 million x only 8-12 bits per pixel (one
color), so size-wise it has a 2-3x size advantage over TIFF at 6mp
x 24bits (three colors). Software to open RAW images must
reconstruct the color interpolations similarly to what is done in
the camera. Since every pixel in a Foveon image has three colors,
a RAW format doesn't give that same automatic 2-3x size reduction
advantage.
