Sigma SD9 Review

Review based on a production Sigma SD9, Firmware Version 1.3.0.1040

Introduction

Foveon caused quite a stir back on the 11th February this year when they announced their revolutionary new X3 sensor. On the same day Sigma announced that the SD9 would be the first digital camera to utilize this sensor, of course it would also be the first digital camera made by Sigma. In time for that announcement we had the first exclusive hands-on with an X3 sensor published in this article. The announcement by Foveon and Sigma was the talk of PMA 2002 and it's fair to say that the show was dominated by talk of X3 and the promise it could bring.

Since that announcement there has been a noticeable period of silence from both Foveon and Sigma which ended just before Photokina when they revealed the relatively low $1799 list price for the body. During Photokina we got our first taste of a pre-production SD9 and were able to post some of the first sample images from the camera.

UPDATE May 2003: Please read my comparison of PhotoPro v1.0 (used for samples provided in this review) and the latest version PhotoPro v1.1 which has several significant image quality improvement features. Click here.

Terminology

Foveon has recently decided to use new terminology to describe its sensor and the final output image, they have requested that I use that terminology in my review (I agreed the new terminology is more logical and easier to understand).

• Photodetector - A physical device used to capture a single color value (for a Bayer camera there is one of these per pixel location, the X3 sensor has three per pixel location)
• Pixel location - A location on the final image which contains full RGB color information

What's so good about this new X3 sensor?

In the example below we are simplifying things by using a 130 x 130 sensor, thus a total of 16,900 pixel locations in the output image. Obviously the same rules apply no matter many pixel locations the final image has.

Traditional "Mosaic" sensors

Digital cameras have for years used what is essentially a monochromatic image sensor with individual color filters over each photodetector (known as the Bayer pattern and normally in a GRGB pattern). This means that each photodetector can only detect one of either red, green or blue light. Note that the green channel has twice the number of photodetectors as red or blue (this is because the majority of luminance information is carried in the green channel). A de-mosaic algorithm then combines the color of neighbouring pixels to reproduce the scene color at that pixel location.

Red 4,225 photodetectors	Green 8,450 photodetectors	Blue 4,225 photodetectors	Output 16,900 pixel locations

Foveon's X3 sensor

*

Foveon's sensor instead of having a single photodetector at each pixel location has a multi-layered design which has the equivalent of a red, green and blue photodetector at each pixel location. This means that instead of having to interpolate the values of neighbouring pixels the X3 sensor 'sees' full colour at individual pixel locations. This is pretty much the 'holy grail' of digital image sensors, it should mean ultra sharp, detailed images with full color single pixel resolution.

* Note that this is a simplified drawing of how the chip works, for more detail see our news article at the time.

Red 16,900 photodetectors	Green 16,900 photodetectors	Blue 16,900 photodetectors	Output 16,900 pixel locations

You can read more about the X3 sensor by following any of the links below:

• Foveon X3 technology overview (11/Feb/02)
• Foveon's revolutionary X3 sensor (11/Feb/02)
• Sigma SD9: Foveon X3 sensor (11/Feb/02)
• Foveon X3 exclusive preview (11/Feb/02)
• Foveon announces Adobe and Microsoft support (23/Feb/02)
• 2002 DIMA Innovative Digital Product Awards (26/Feb/02)
• When is a pixel not a pixel? When it's three (6/Mar/02)
  

Review notes

JPEG notes

Unless you didn't yet know the SD9 doesn't provide in-camera JPEG production, instead it writes RAW (Foveon X3F) files. This presents problems in publishing a review like this with samples. Ideally we would provide uncompressed TIFF files (from the provided RAW converter - Sigma Photo Pro) in this situation but obviously that's impractical from a download and bandwidth point of view. The next solution was simply to output JPEG from Photo Pro.

However, the initial release of Photo Pro appears to subsample chroma (color) information at a 2x2 size, this is acceptable for high compression but means that detail is lost when low compression, high quality JPEG's are required. Foveon have assured me that they are working on this issue and will have an updated version of Photo Pro available soon.

Therefore all JPEG samples provided here were first output from Photo Pro as 8-bit TIFF (sRGB color space) and then converted to a 95% or 99% quality JPEG via an Adobe ImageReady Droplet. This produces very high quality JPEG's with no visible artifacts, considerably better quality in fact that the normal best quality available from any other digital SLR.

Any comparisons made to other cameras in this review will be carried out using that cameras RAW mode, output as TIFF and converted to JPEG using the same procedure.

If you're new to digital photography you may wish to read the Digital Photography Glossary before diving into this article (it may help you understand some of the terms used).

Conclusion / Recommendation / Ratings are based on the opinion of the reviewer, you should read the ENTIRE review before coming to your own conclusions. Images which can be viewed at a larger size have a small magnifying glass icon in the bottom right corner of the image, clicking on the image will display a larger (typically VGA) image in a new window. To navigate the review simply use the next / previous page buttons, to jump to a particular section either pick the section from the drop down or select it from the navigation bar at the top. DPReview calibrate their monitors using Color Vision OptiCal at the (fairly well accepted) PC normal gamma 2.2, this means that on our monitors we can make out the difference between all of the (computer generated) grayscale blocks below. We recommend to make the most of this review you should be able to see the difference (at least) between X,Y and Z and ideally A,B and C.

This article is Copyright 2002 Phil Askey and the review in part or in whole may NOT be reproduced in any electronic or printed medium without prior permission from the author. For information on reproducing any part of this review (or any images) please contact: Phil Askey.