Digital BW, The Print

>> On the sensor itself,
>> there are 6.3 million real pixels, but each pixel only records one colour
>> (red, green or blue).
>
>Well, no.  There are 6.3 million real SENSORS, not pixels.  A COLOR pixel,
>as used in any imaging program, consists of ALL THREE COLOR COMPONENTS,
red,
>green AND blue, not just one of them.
>
I meant the same as you, but yes, sensor is a better term. It's less
confusing.

>> The new Foveon chip records red, green AND blue values at every
>> pixel on its
>> surface, so presumably it will give better quality than a Bayer sensor at
>> the same resolution.
>
>It's somewhat the same as just taking a group of four of the Bayer pattern
>sensors and calling them a pixel ;-)
>
Similar, but not the same. Each bayer pattern sensor is recording light
intensity (for one colour) at a slightly different location in the grid.
It's just that each sensor only has information for one colour channel.
(BTW, I've often wondered why nobody ever designed a bayer filter grid that
could somehow be moved out of the way so that each pixel recorded just the
light intensity, letting the user make full-resolution monochrome images.)

I read an analysis of the quality of images derived from Foveon's new
sensor, and the author calculated that they are equivalent (in perceived
quality) to a Bayer pattern sensor with something like 1.4 or 1.6 times the
resolution. It's an improvement, but not as great as the 4x resolution
improvement you'd expect at first glance. Sorry not to be able to back this
up with a reference, but I can't remember where I read the analysis.

Another paper I read was a study of the lp/mm it is possible to capture on
35mm film. The author found that lenses, use of consumer film, non-use of
tripods/shutter releases/mirror lock-up, use of autofocus, filters placed on
the lens, etc limits most photographers to a quality of around 50 lp/mm.
That equates to 100dpi on a sensor 36mm x 24mm, or a 3600x2400 pixel image
(just over 8Mp). Using all the right techniques and top-of-the-line slide
film it is possible to achieve 80 lp/mm, which equates to a 22 Mp digital
image. Of course, noise characteristics differ (random grain vs.
evenly-spaced pixels), so percieved quality may not match up like the maths
does.

In the end, though, the mathematics doesn't matter. It's what the user likes
best that matters.


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