Foveon Sensors and Autochromes

You may have heard about Sigma’s newly released high-end camera called the SD1. If so, what you probably heard about was the price tag. A list price of $9,700 seems pretty high, but there’s some pretty neat technology hiding in it. It uses the latest version of something called the Foveon sensor.

The important difference between this sensor and the ones used in other digital cameras lies in how the color information is collected. The basic sensing element that these sensors are built out of doesn’t capture any color information. To collect color information, you need to pass the light through filters before it reaches the sensor.

Film emulsion has the same issue of course. Early pictures were black-and-white because nobody had figured out a good way to capture the color information. One of the very first color processes was something called an Autochrome. In an Autochrome, the film is covered with tiny starch granules. These are dyed three different colors (not RGB, BTW). The light passes through the colored granules before hitting the emulsion. This means that the exposure of the emulsion under the different colored granules contains information about the color of the light which exposed the film. If the incoming light was blue, then the emulsion under the bluish granules will be more exposed than the emulsion under the other color granules because they would have blocked the bluish light. The starch granules remain on the film after it is processed. This means that when you look at an Autochrome, you’re seeing the emulsion through millions of tiny color filters which are exactly aligned with the filters which were used when the film was exposed.

Autochromes had a number of problems, and better techniques were invented for color photography. Most of the other techniques involve stacking three layers of emulsion with filters between them. This gets rid of the random pattern which the starch granules made in an Autochrome. The only pattern in a chromogenic film is the one created by the tiny silver crystals.

It turns out that the difference between the sensor in the SD1 and the sensor in other digital cameras is very similar to the difference between Autochromes and other color film processes. Most digital cameras collect color information by covering different sensor elements with different colored filters. The most common layout is something called a Bayer filter.

In a way, this is similar to how an Autochrome worked. The colored filters are imposing a pattern on the incoming light. When you impose this kind of pattern, there are some interesting questions about whether it should be very regular (like the Bayer filter) or very random (like the starch granules). I looked into this a bit when I was at Applicon in the early days of ink-jet printers. It turns out that neither extreme is optimal. It appears that the best approach is usually something like Poisson disk sampling, or the one which Robert Ulichney described in his paper Dithering with Blue Noise.

The Foveon, on the other hand, uses a system which is more like “normal” chromogenic color film. It consists of three stacked layers of sensors which are sensitive to different color light.

Is the Foveon approach better? Will the Bayer filter disappear like the Autochromes did? I don’t know, but it is very interesting to see this echo of technology war from a century ago, isn’t it?