Steve's Digicams Forums

Steve's Digicams Forums (
-   General Discussion (
-   -   Sensor types and layout question (

sidneylopsides Mar 30, 2005 4:43 PM

From what I know:
A Bayer filter array CCD has RGB colour filters over the pixels, with twice as many greens ones. A 4Mp sensor would have 1m red, 1m blue and 2m green. Each colour channel uses Bayer interpolation to fill in the gaps and the colour channels are combined for the final image.

A SuperCCD is similar, but with octagonal photo sites.

A Foveon type has an red, green and blue sensor at each pixel location, so a 3Mp Foveon produces a 1Mp image, but as there is no interpolation the image is sharper.

Is this correct? Someone is telling me that a standard CCD has pixels made of three subpixels, like a monitor.

Thanks in advance

Lin Evans Mar 30, 2005 6:13 PM

Pretty much correct except for the Foveon. If you're referring to the Foveon as used in the Sigma SD9/SD10 it's a 10.3 megapixel sensor which creates a 3.4 megapixelfile matrix. But this 3.4 megapixel file has the optical "resolution" of a six to eight megapixel bayer sensor and because it has no antialiasing filter produces extremely sharp images. It also resolves to the single pixel level - something not achieved by bayer designs.

The confusion lies in the definition of the term "pixel." Typically, in all except Foveon technology a "pixel" and a "photosite" are interchangeable terms because the final file produced has essentially the same number of display pixels as the sensor had photosites. In the bayer processed sample, colors at a given locus (place in the environment as sensed by the photosite) are indirectly determined by a process loosely known as "interpolation." More correctly in this instance it should be called extrapolation. Luminance values (brightness) are detected by each photosite but colors are extrapolated by comparing the values of adjacent samples and both an antialiasing and bayer color filter are used to determine the final pixel values which are used in the display or print.

As "obtuse" as this may seem, a long period of trial, error, iterative adjustment and experience have made this a quite accurate procedure. With the Foveon X3 processor, rather than a single layer, two dimensional processor, a three dimensional processor containing three deep photosites directly senses RGB values. This means that each display pixel value comes from a combination of values as sensed by each of the three separate photosites ateach locus. The essence is that the values are in theory more accurate so that each of the 3.4 million pixels in the file matrix has a more "precise" value which lends itself to producing an image with less "smear". In practice the Sigma/Foveon image has fewer pixels in the transition from edges of detail to relative background (edge roll-off) and this makes an image which has greater sharpness with less need for USM (Unsharp Mask). USM is a means for making an image look sharper without actually being sharper by increasing relative contrast at the edges of detail features. This enhanced edge contrast draws the eye to that region - sort of like outlining letters with contrasting color to make them stand out.

The Super CCD as used in the Fuji series of cameas uses a beehive arrangement raher than a square arrangement of photosensos which apparently increases the resolution along diagonal detail. The essence is that some overall increase in measured resolution is achieved simply becauseof "statistical averages" innature there are a number of detail features which fall outside of relative 90/180 degree positioning. An increase in the resolution at these positions results in an overall sharper appearance. On the down side, especially for consumer level Fuji cameras, there is quite often a rather grainy appearance to the Super CCD images. Not exactly increased "noise," but similar in appearance. In print this is rarely an issue but viewed at 100% on screen it's sometimes less than aesthetically beautiful.

The bayer process is generally considered to be better overall in the sense that very smooth appearance of images are possible. Even at high ISO amplifications noise is less apparent than with the Foveon approach. But for pure sharpness and especially for enlargeability my Sigma SD10's output is equal or superior to my Canon EOS-1D Mark II (eight megapixel).

These are simply different approaches to sampling the environment and some are better for some purposes than others. Like any tool, the photographer must choose the approach which works best for the job at hand. Sometimes it's one of my Canon dSLR's, sometimes it's my Sigma and sometimes it's one of my Fuji's.


sidneylopsides Mar 30, 2005 6:45 PM

Thanks :) That is how I thought it worked, my brief descriptions weren't as accurate though.
Just to check, a 4Mp RGB Bayer sensor would have 1m red, 1m blue and 2m green pixels?

All times are GMT -5. The time now is 8:45 AM.