Making Sense Of Sensors; Interpolation Before And After The Fact Page 2

Summing It Up
What does all this prove? Basically, that if your reference standard is an unmanipulated 13x19" print, a practical comparison test of the Foveon X3 sensor suggests that it can deliver imaging performance--especially in the areas of sharpness and color detail--that's comparable to the sensors in two of the leading D-SLRs in the 10- to 12-megapixel class. And while the Sigma SD14 and the Foveon X3 sensor didn't "win" either of the face-offs, the fact that it can even play in the same ballpark with the big boys means that it deserves a good deal of respect.

The issue here is what happens in terms of interpolation, or upping image size after the exposure is made. There is no question that all image-processing systems in digital cameras interpolate information--that's what makes digital imaging work. And that's what has created controversy about the Foveon sensor--the fact that when you open the Image Size dialog box the file size is 13MB, which makes photographers think they have a 4.6-megapixel sensor and not the 13.8-megapixel sensor that Foveon implies with their three-layer architecture. But when interpolated up, even using Adobe's method in CS3, results are on a par, in terms of sharpness and clarity, with the 10+ megapixel sensors using the more "conventional" Bayer pattern. Does this speak more to the effectiveness of interpolation algorithms than to the competition among sensors? That's what makes comparing sensors such an interesting challenge. In short, our tests show that there is more than one way to make a good image sensor, and having a choice benefits all. It will be fascinating to see what Foveon comes up with next--a DX-sized three-tier CMOS sensor with 6 megapixels per layer might well put them over the top.

Many thanks to photographer and digital printing expert Joseph Meehan, who printed all the test pictures for this article. Without his enthusiastic and cheerful support, executing this piece would not have been possible.

Still Life
Canon EOS 5D
Nikon D200
Brilliant, moderately oversaturated color with excellent detail at the point of optimum focus--note well defined thread on spool, lined pattern in fishing line at left of spool, surface texture in orange lure with eye and "feelers."
Very accurate, less saturated color than Canon; looks very natural. Details at point of focus--thread on spool, patterned fishing line, orange lure--are well delineated, but not quite as sharp as Canon, but other details such as speckled lure with pink "tail," texture in black fishing reel bag and reel itself are noticeably sharper due to greater depth of field at 1.5x longer shooting distance than Canon.
Sigma SD14
Good color saturation, but color accuracy is affected by slight overall greenish cast, visible in fishing reel, lower right, blue area lower left, slight greenish cast in yellow and purple material, upper right, which looks blue. Overall detail rendition and sharpness are outstanding--on a par with Canon's performance at the point of focus.
Conclusion: We'd call this a toss-up. If we were forced to pick one winner, it would be the Nikon print. As for the Sigma SD14, its performance in terms of detail sharpness and rendition is once again impressive. Had it not been for the slight greenish cast, it may well have edged out at least one of its rivals. But as we said before, color correction was not part of our test protocol.
Photos © 2007, Jason Schneider, All Rights Reserved

Sensor Face-Off: Upsides And DownsidesHere's a brief look at what differentiates the two contending sensor types:
The Bayer-pattern sensor, originally developed by a scientist at Kodak, detects color using a defined, repeating checkerboard pattern of green, blue, and red micro-filters placed on top of the light-sensitive silicon layer. These tiny filter squares are added in a specific ratio of colors--50 percent green, 25 percent blue, and 25 percent red--with one filter placed at each pixel location. As a result of this single-layer structure, a Bayer-pattern CCD or CMOS sensor provides a very large number of pixel locations per square inch. Today's Bayer-pattern sensors deliver excellent sharpness and color accuracy as a result of constant improvements in sensor manufacturing technology and image-processing software. (Editor's Note: Kodak has recently released news about a Bayer-pattern derivative, a so-called "panchromatic pixel"-type filter in which no color is overlain in certain areas, although the familiar RGB array remains over most of the sensor. Their claim is increased light sensitivity (up to 2x) and less noise.)

The structure of the Foveon X3 CMOS image sensor is quite different, more closely resembling that of color film, with three separate light-sensitive layers aligned directly above one another. By placing a separate pixel sensor at depths corresponding to the highest density of electrons generated by blue, green, and red light, the three-layer X3 sensor can capture full color data in blue, green, and red at each pixel location. Having better, more complete color information at each point means that very fine color details can be rendered more accurately, enhancing overall performance. The downside is that the X3 sensor has fewer pixel locations per square inch, which results in a smaller file size. In addition, the X3 sensor, unlike the Bayer, employs wide-band color filters that allow some light of the "wrong" color to be captured in each layer. This information must be subtracted during image processing.