Similar to the human eye image sensors consists of small light
sensitive receptors = photodiodes.
These photodiodes are sensitive to only one
That's the well known RGB schema also used in (classic) TV screens.
Now what's the difference then between a photodiode and
a pixel ?
There's a lot of confusion out there and some don't differentiate at
all between the two. Personally I prefer to think of the following
definition: A pixel can be calculated based on data obtained from three
or more photodiodes (R + B + G + ...). Have a look at the following
explanation for details.
"Bayer"-type Image sensors
Image sensors with Bayer interpolation are currently the most popular
variant of the species. "Bayer" sensors are layered out in (overlapping)
clusters of 4 photodiodes. A cluster is based on 2 green-, 1 red- and 1
blue-sensitive photodiode (to be precise these are "plain" photodiodes
with color filters). An individual pixel can be "Bayer-interpolated"
based on data from 4 photodiodes (obtaining the necessary RGB
information). Now why's that an interpolation ? Looking at the
illustration below you may notice that the 2 right hand photodiodes of
Pixel #1 are shared with Pixel #2. Therefore there's a certain amount
of redundancy within the system. So while you may interpolate e.g. 6
mio. pixels ((xmax-1)
* (ymax-1) to be nit-picky) from 6 mio.
photodiodes the effective resolution is actually a little lower than
Lately a few variations to this theme emerged:
Sony replaced one of the 2 green photodiodes of the 4-diode cluster
with a new photodiode featuring an emerald filter (therefore RGBE).
Similar to what we've seen on the inkjet printer market the idea is to
improve color accuracy and according to the first reviews this new
approach seems to work out pretty well.
Another interesting variation has been developed by Fuji:
Current digital cameras still suffer
from a limited dynamic range (=range between deep black and bright
white) resulting in a
very limited ability to resolve details in shadows or highlights. Fuji
found a workaround for this problem by adding a secondary (smaller)
area to a photodiode with a lower light sensitivity. This trick is supposed to provide up to 2 extra
stops in dynamic range compared to conventional photodiodes.
Looking at the illustration below you also may observe another
difference compared to the classic layout - the photodiodes have an
octagonal shape with an honeycomb layout
resulting in a potentially higher density because the photodiodes of
two neighbouring rows can overlap a little. Due to this special layout
Fuji S3 Pro interpolates their max. output to 12 megapixels based on 6
mio. photodiodes. While this may be a little optimistic the "Super CCD"
layout seems to be indeed superior regarding
the effective resolution compared to Bayer-sensors with the same number
CDD" (SR) layout
The next image illustrates the vertical structure of Fuji's Super CCD (HR) sensor -
apart from the Fuji-style layout and form of the photodiode the layers
are typical for image sensors (incl. Foveon) used in D-SLRs. The
surface layer is made of microlenses. Why's that ? The photodiodes are
slightly recessed and if light rays don't hit the photodiodes in a
perpendicular angle, and that's typical for the image edges, they
produce slight shadows resulting in a lower exposure. On a macro level
(the chip) the exposure may be uneven from the center towards the
edges. Microlenses are meant to correct this problem (and recently
manufacturers started to adjust their lens designs due to this issue).
The 2nd layer consists of the photodiodes as described above whereas the
third layer cares about the actual data transfer.
The "Foveon" sensor
As mentioned above Bayer-type sensors are, by a vast margin, the most
around but there's a little company rivaling the big ones - "Foveon".
Foveon took advantage of an effect well known to B&W photographers
(and also scuba divers): Using filters you can block one color whereas
colors remain unaffected. So they simply stacked three semi-transparent
layers of photodiodes rather than spreading them like on Bayer-type
individual Foveon pixel
(made of three layered Photodiodes)
This approach has some advantages and a few drawbacks. On the up side
you can list that an individual pixel
location produces one "true"
color& luminosity information - there's no interpolation needed and
yes, images tend to be quite a bit sharper compared to conventional
Bayer sensors of the
same pixel resolution (output). As a nice side effect the photodiodes
COULD also be much (3x) bigger
thus resulting in less sensor noise - POTENTIALLY that is which is not
the case as of now.
On the downside you need a
LOT (3x) more pixels in order to achieve the same
output-/pixel-resolution compared to Bayer-sensors. The Sigma SD10 has
e.g. 10.2 mio. photodiodes - that's more compared to an Canon EOS 10D
or Nikon D100 with just 6 mio. photodiodes and almost as much as on an
Canon EOS 1Ds with its 11mio photodiodes. However, as these pixels are
stacked the effective output resolution is really just
3.4 mio. pixels (10.2 mio. / 3 layers) and that's not too impressive
from the perspective
of today's Bayer sensors.
So while the Foveon solution has a much better potential on the long
run they've a marketing problem as of now - a 6 megapixel Canon EOS 10D
just sounds more sexy than a similar-priced 3.4 megapixel Sigma SD10.