Foveon (was CMOS vs CCD digital)

From: Etienne Garbaux ^lt;>
Date: 02/13/04-10:00:34 PM Z
Message-id: <p05210601bc532c9771ca@[]>


>an interesting concept-I've been following the Foveon chip since it was
>announced. I like the idea that the chip records all three colors (RGB) at
>each sensor site-this should give much better sharpness in images and
>perhaps sharper black and white too. Other technologies use a staggered
>array of sensors for seperate recording of R, G, B. Then they interpolate
>the colors in between-which is where you get some color shifts and loss of


> Actually, you need to remember that most digital cameras use 4 sensors in a
> group (two green, a blue and a red). So while the Foveon chip is a true 3.4
> Mp chip, most other cameras will need 4 times the megapixels to capture the
> same detail (i.e. 13.6 Mp). In practice there are some tricks which can be
> used to fool the viewer into thinking that the non-foveon chips have higher
> resolution than they do.


> I haven't been too impressed with the Foveon. Better "theory"
> than execution? Perhaps the displacement of the RGB sensors in depth is
> every bit as bad as the displacement in width on a conventional chip?

Foveon likes to analogize their three-layer sensor with tripack
(three-layer) color film. The thing that Foveon doesn't say is that their
sensor really doesn't work the same way. In the old days, color
separations were made by making three separate exposures, one each through
special Red, Green, and Blue color separation filters. The R filter lets
through the longest wavelengths, blocking the middle (G) and shortest (B)
wavelengths. Similarly, G lets through the mid-wavelengths and B the
shortest, each blocking the other two.

Tripack film makes use of the fact that silver-gelatin emulsions can be
made sensitive to blue, blue and green, or blue, green, and red
wavelengths. The first layer (toward the lens) is un[color]sensitized
(that is, sensitive to blue light only, like any silver halide emulsion
without organic color sensitizers). Thus, even with light of all
wavelengths falling on the film, the first layer RESPONDS only to blue
light -- as if it had a blue filter in front of it, but it doesn't need
one. Under the blue layer is a yellow filter, which removes the blue
wavelengths and allows the G + R through. The second emulsion layer is
orthochromatic (sensitized to blue and green wavelengths, but not red).
Thus, it forms an image only with green light -- the B has been removed by
the yellow filter, and the emulsion is not sensitive to red. Under the
second emulsion layer, there is a red or magenta filter, which lets only
red wavelengths through to the third emulsion, which is panchromatic
(sensitive to all visible wavelengths, like most common B&W film). This
layer forms an image only with red light, because all other wavelengths
have been filtered out. [This paragraph describes basic, old-technology
tripack color film. Modern color film has lots of enhancements and some of
the foregoing is no longer strictly accurate.]

Now, the Foveon. Unlike color film, the sensitivity of each layer is not
adjustable separately from the filtering -- each layer of the sensor
responds to all of the wavelengths reaching it. So, the top layer, which
has no filtration in front of it, responds to all wavelengths -- it
produces the same image that a panchromatic B/W film would produce.
Doesn't sound like a very good start toward a color layer, does it? For
that, we'll need to SUBTRACT the image from the next layer (see below).
Some distance below the top sensor is a second sensor. Because silicon
attenuates light preferentially -- shorter wavelengths more than longer
ones -- at some depth, an effect something like a yellow filter is
obtained. By placing the second sensor at that depth, it yields a
minus-blue image -- that is, R + G. Now, by SUBTRACTING that image from
the image from the top layer, we have (B + G + R) - (R + G) = B. Two
sources of error can make this blue image less than ideal. First, the
color-preferential attenuation of light in silicon does not produce
anything like the sharp band cutoffs that the special separation filters
give, and only approximates a useful filter characteristic for color
photography. Second, the blue image is composed by combining two sensor
images, thus doubling the noise.

Some disance below the second sensor is a third, placed at a depth where
the attenuation profile of the silicon substrate approximates the action of
a red filter. This sensor image is the red image. The same comment as
above about the approximate nature of the filtration applies. By
SUBTRACTING this image from the image from the second layer, we get a green
image (R + G) - R = G. The same noise comment from above applies.

So, after some processing, we get three approximate R, G, and B images that
we can expect to be somewhat noisier than their counterparts from a
Bayer-system sensor (everything but the Foveon, at the moment).

There is one other possible source of error in the Foveon. Because the
three layers are at different depths in the substrate, in principle there
must be some focus error from one layer to another. To know how much of a
practical factor this is, we would need to know just how deep the second
and third layers are, which as far as I know Foveon has not stated.

Both systems have their limitations. Photographers who have done
three-color work using filters other than the special R-G-B separation set
have experienced some of the what Foveon must contend with to get
reasonably accurate color.

Best regards,

Received on Fri Feb 13 22:01:22 2004

This archive was generated by hypermail 2.1.8 : 03/02/04-11:35:08 AM Z CST