Foveon claims that there is no color filter array, which is required in other sensor layouts. True, but the substrate layers are the equivalent of color filters.
Photo Diode Receptors
Foveon supporters claim that there is no need for demosaicing because each pixel contains all the light for all colors. Of course when they want advertised pixel count, each Foveon site is counted as three pixels (RGB). For an equivalent Bayer layout, each site is counted as one pixel for four sensors (RGGB). The size of the Sigma SD10 sensor site is about 9.1 um (2783 PPI). The size of the Nikon D100 sensor site is about 7.7 um (3333 PPI). The SD10 has a 3.4 MP image. The D100 is a 6.3 MP image because it has smaller sites and a larger image area. The smaller SD10 image plane actually makes it more susceptible to diffusion.
Each Foveon receptor sees white light. The blue receptor records only blue. The green receptor records green and blue. The red receptor records all three wavelengths. Red equals red minus green and blue. Green equals green minus blue. Note the placement of the voltage meters in the image above. Each meter is reading across a different p-type and n-type junction. Each meter is recording the voltage (intensity) for one set of wavelengths.
The size of the site is not a direct function of the layout. The layout of the site determines the number and shape of the tiles. The individual receptors do not see the same light rays from the same source so they are subject to the same light spread function as a Bayer pattern.
Using a Bayer pattern where it seems that each Bayer element is the same size as a Foveon site, it is demonstrated that Bayer needs demosaicing (softening) while Foveon does not. A demosaicing technique can be employed to increase the effective resolution. It is useful for very low-resolution sensors. It is unnecessary and usually harmful for most high resolution sensors.
Color film has a layered layout like Foveon. It is also tiled like Bayer except that the layout of the tiles is random rather than patterned. I would describe the Foveon tiles as variable sized containers. It would seem reasonable that the surface areas at each sensor should be approximately equal.
The wavelength filtering, electrical cross talk, and diffusion of light all affect the "color fidelity" of the sensor. The result is a sensitivity curve for each channel that shows a similar correlation for Foveon as those comparing different Bayer sensors from different cameras. The Foveon curves are smoother and generally all three are broader. See below.
These curves must be applied to the data stored for each sensor photo diode, receptor, element, or channel. Use whatever term you like. The non-linearity of the sensors has to be normalized to the linearity and range of the spectrum. This is the only way to arrive at the correct RGB values for yellow (for example) at the 570 nanometers wavelength. Obviously, the algorithms vary among the different makes and models of the sensors. Do not confuse this with the human visual response.
The same sensor data processing that is applicable to other sensor designs is just as applicable to Foveon sensors.
That is just my two cents. These are my own observations and opinions. They do not support or reject the claims of any manufacturer. I hope you also gained some new insight from this article. If you have any comments, or suggestions, I would welcome your input. Please send me an Email.
Most of the images here were snipped from: http://www.foveon.com/docs/Century.pdf. The article is "Eyeing the Camera: Into the Next Century" by Richard F. Lyon and Paul M. Hubel, Foveon, Inc.
Rags Int., Inc.
204 Trailwood Drive
Euless, TX 76039
September 6, 2004
This page last updated on: Thursday April 12 2007
You are visitor number 15,000 since 09/07/04