In article < >, Georgette
Preddy <> writes
>Prometheus <Prometheus@127.0.0.1> wrote in message
>news:<>...
>> In article < >, Georgette
>> Preddy <> writes
>> >Pixel count can be inflated to any number using digital interpolation
>> >after the shutter closes.
>>
>> Rather like the smeg SD9/10 which detects light at 3.2 million loci and
>> the makes up another 9.8million with its companion software.
>
>Fine, if you count each color channel in isolation, the 1D Mk II only
>has 2M measured locations.
>
>The SD9 (3.5M) has more than the 10D (1.5M) and 1D Mk II (2M),
>combined.
The sensor in the SD10 has 2268 columns x 1512 rows (as stated at
http://www.sigma-photo.com/Html/news/news_sd9_fs.htm), or from
http://www.foveon.com/prod_f7.html it has a pixel pitch of 9.12ĉm
centre-to-centre spacing of pixel locations with an effective area of
20.7 mm x 13.8 mm; both of which give a total of 3.43 million pixels in
the image plain. The sensor only detects light normal to its surface at
the intersections of the rows and columns, it has no information about
the light falling on the interstice. The SD10 gets 3.43 million pixels
from 10.3 million sensor elements in 3.43 million locations on its
sensor. The 10D gets 6.3 million pixels from 6.3 million (3072 columns x
2048 rows) distinct locations in the image plain. Neither interpolates
pixels, by default. They both interpolate colour, although the Foveon
attempts to measure full colour at each pixel. See
http://www.siliconimaging.com/RGB%20Bayer.htm for an explanation of the
Bayer process and there is a Bayer algorithm
http://www.tortuga.comau/products/info/cfa2bmp.html from these you will
see that the luminosity at each detector is used and guessed, unlike the
SD10 which has to guess.
A Bayer sensor outputs the exact (as least as any measurement can)
luminance for green on a green sensor, the same goes for blue and red.
The other two luminances are computed from the 8 surrounding pixels.
This computation assumes that the chrominance varies slowly. This
assumption can be made because our own seeing mechanism is not all that
good at detecting chrominance changes, i.e. if there is a "mistake" in
the actual luminance then we will not see it.
The Foveon sensor also measures an approximation to reality, since its
colour response is so far from matching the human perception of colour.
The colour you see in any Foveon image is at least as much a
"computer-generated artefact" as a Bayer image. In addition, the low
spatial resolution of the Foveon sensor irreparably loses spatial
information that would have been captured by a Bayer sensor, aliasing it
to false detail because there's no anti-aliasing filter.
SPP allows you to manipulate the data whilst it was still in the camera
but as such they are not straight from the sensor, in fact since you
stated that they are straight out of SPP we must conclude that you used
SPP to multiply the number of columns and rows thereby inventing values
where there are no detectors. The result of this manipulation tells us
nothing about the number of pixels the SD9 has. Sigma do tell us it has
2268 columns x 1512 rows (as stated at
http://www.sigma-photo.com/Html/news/news_sd9_fs.htm), any image having
more has been manipulated by external software.
Let me try to simplify it; SPP might "optionally output 14MP images" but
the sensor only resolves 3421656 unique points in the image plain,
anything more than this is a software cheat by the user! Or even
simpler: more than 3421656 is dishonest, less other than a crop is a
disservice. Simpler still - you dissemble.
Your 14MP SD9 uses luminance and chroma extrapolation to provide 10
million pseudo pixels where it can not detect light, the 10D uses chroma
interpolation to provide colour at the 6.3 million where it does detect
the luminance; not even remotely the same thing.
--
Ian G8ILZ
Similar Threads
