Remember the old tape format battles in the days before DVD? There was VHS and Betamax. With great regret, the poorer format devoured a technically better system - and we were consigned to VHS until at long last Dixons managed to ditch it! Now there are two types of sensors for digital cameras - and I'm not referring to CCD versus CMOS. There are cameras with Foveon sensors, and there are those that use a single layer chip known also as a 'mosaic sensor' or 'colour filter array'. The purpose of this article is to demonstate that, at the time of writing, non-Foveon sensors will always produce a poorer final image. Just in case you might imagine that I'm highly prejudiced in my conclusions (before you've read the rest of the article), I own and use two DSLRs - the Sigma SD9 and a Nikon D70. Both cameras have their strengths and weaknesses - but the D70 is 'state of the art' and up-to-date with a wide ISO range, minimal noise, better low-light focussing, but still does not produce images with the same detail as the SD9 - it is also prone to moiré artefacts - a typical by-product of using a single layer sensor. I'm not suggesting that the SD9 sensor is 100% perfect - some of the downsides include a very limited upper ISO setting which has been resolved in the SD10, and over blue skies.

Before we make a start, I am really curious as to why the Foveon sensors have only found their way into so few cameras - two Sigma units and a Polaroid? Why doesn't it appear in cameras from Canon, Nikon, Olympus etc? My own pet theory is that the larger camera manufacturers are stuck with mosaic sensors because they need to overcome the limitations by using their own processing solutions. One interesting comment came from DPReview's analysis of the SD10;-

'Looking back at my SD9 review I stand by much of what I said about the X3 sensor being 'the first step in what must be seen as a revolution in digital photography'. Unfortunately the stranglehold the Bayer sensor and those mega-corporations who make them have on the market has ensured that we haven't yet seen the X3 sensor in another digital SLR.'

We are all familiar (I hope) with the physical differences of a standard sensor and the Foveon. The schematic above shows the essential characteristics. The Foveon chip captures colour by exploiting the fact that blue light waves are shorter than green light waves, which in turn are shorter than red ones. It uses three layers of photosensors. Note that it captures 100% of RGB whereas the mosaic sensor captures only 25% of Blue and Red, with 50% of Green (because the human eye is most sensitive to green light). I was most interested in a reply to a thread which I started in the Sigma forum of dpreview.com. I had asked the question because of an issue raised in Bruce Fraser's book on 'Real World Camera Raw with Adobe Photoshop CS' (see footnote). In it, he comments; "But the sensors in the array, whether CCD or CMOS, just count photons - they produce a charge proportional to the amount of light they receive - without recording any color information". So a RAW file simply records the luminance value for each pixel - it is a greyscale image containing colour information. The question I had posed was to ask if I was correct in assuming that X3F RAW files contained colour information embedded in them? The following reply from 'SigmaSD9' is so illuminating, that I will reprint it entirely:-

"The answer to your question is yes. Bayer sensors are monochrome. The reason is that they sense monochrome data at every photosite. The sensor is just a grayscale device with a colored plastic mosaic glued in front of it. The way you get color out of a Bayer sensor is by digitally interpolating the missing color channels at each monochrome photosite, using neighboring data. The magic of digital lets you borrow the color from the pixel next door. The end result of the borrowing is that the recorded pixels are no longer optically discrete entities, which is why it is incorrect to think that an "8MP" Bayer can produce 8M optically discrete full color pixels. It can't. In fact it can't even come close. Foveon colors aren't the result of digital interpolation, but are rather optical, as each photosite has complete RGB data. No borrowing things sensed elsewhere on the sensor are required. So every pixel in the output image is an optically discrete entity, it's not an upscaled image derived from lesser core daya. Although the RAW files themselves don't really drive that, the physical sensor geometry does".

The simple fact is that mosaic sensors simply cannot achieve their quoted resolution. Mac User (26th November 2004) makes much the same point. In an article "Technical Briefing: How Digital Cameras Convert to Digital", it states that it takes a total of four pixels (2 x Green, one each of Red and Blue) to create one full-colour pixel, and thus it seems impossible for cameras to achieve their advertised resolutions. It continues; "This approach isn't without its limitations. Despite some advanced signal processing, detail is by necessity compromised. Colours and detail are created by sophisticated interpolation, and aren't absolute representations of reality. Foveon's X3 chip is more advanced in this respect".

Colour filter array interpolation (demosaicing) is the process of giving each pixel a full set of RGB values. In fact there are many methods of demosaicing - some more prone to generating their own artefacts. If you then add the JPEG rendering process found in other cameras, whereby up to 75% of the available data is compressed or thrown away, you begin to undestand the resulting mess. One method of applying demosaicing is to work out how similar the surrounding pixels are to whichever pixel is being examined. Looking at the grid (above), to find the red value for sensor G13, we check the nearby red sensors. If they are very similar, it might be a safe bet that the value of red at G13 is the same as R12 and R14. Perhaps the most polite thing you could say about Bayer array sensors is that they attempt to guess colour.

Let's return to Mac User again; "One alternative technology of note is Foveon's X3 chip, which mimics film-based photography much more closely. Since the demosaicing algorithms aren't needed, images produced from X3 chips are theoretically much less susceptible to moiré artefacts". Moiré patterns are caused if the subject has more detail than the resolution of the camera (put crudely), and some cameras offer anti-aliasing filters to reduce the effects - thereby further reducing image sharpness!

If this isn't enough to bring out the Champagne, Adobe produces some truly excellent resources for educators. The Adobe Digital Video Curriculum Guide - Module 2 (Digital Still Photography: Issues and Techniques), has an entire section on Foveon sensors. Their conclusion is that "CCD and CMOS technology pales in comparison to the image clarity of Foveon.... the result is sharper images and more accurate color reproduction". (see footnote). Luminous Landscape website includes a powerful testimony to the major advantage of shooting both on Foveon sensor-based cameras and in RAW mode rather than JPEG. The article can be found here and concludes, 'By itself no sensor can record colour. All imaging chips except that from Foveon... use what is called Bayer Matrix or colour filter array'.

In a fascinating experiment, Mike Chaney has attempted to compare a 3.4MP image from a Foveon sensor to what we would expect from a 3.4MP Bayer sensor. You can find out the full details here. But his conclusions are quite startling:-

• Bayer sensors have a very significantly reduced resolution when resolving detail comprised of mainly red/blue primaries, such as a red sports car with with black pin stripes, a blue sweater with red lettering, red soda can with black lettering etc. In these cases, resolution of the Bayer sensor is reduced to less than 1/4 of its 'image' resolution! Black and white details will show the highest resolving power on a Bayer sensor, while saturated color detail will vary greatly. A Foveon sensor is much more consistent, resolving near the full resolution of the images for each color combination.
• Bayer sensors will produce images that are softer and less detailed due to the "smoothing" needed to eliminate artefacts and color distortions.
• Bayer sensors tend to omit color information when sampling high frequency detail. If you look at the picture of a tree that has many small branches with a brick wall behind it (for example), you will see many of the smaller branches "morph" into the colour of the bricks in the background. This is because the branches are not wide enough to cover the multiple pixels needed to derive accurate color information on a Bayer sensor. Full color sensors eliminate this problem.
• Beyer sensors tend to produce color moiré on high frequency detail.

The bottom line

Well, you bought an SD9 or SD10, so you know it is better! I hope this article will show you why. In a word (or more), the X3 sensor captures true colour, without relying on colour interpolation, nor AA filters to mask the shortcomings of lesser technology. While not strictly a sensor hardware issue, it is always possible that Foveon and Sigma will be able to offer firmware enhancements in the future. We've already seen updated versions of Photo Pro which are able to extract even more data (for example X3 fill light) from the information drawn from the sensor.

In this article, I didn't want to simply present my own understanding of the technical issues, but to point to other resources where the superiority of the X3 sensor has been recognised. So, I leave the final word to a testbench report of the SD9 from Digital Output. It concluded; "The pictures from this camera are uncommonly sharp. We'd say that pixel-for-pixel, the X3 imaging sensor is the sharpest on the market. One reason for this remarkable sharpness - where very small items such as street signs in the distant background are easy to read - is that the blurring filter used in virtually all other digital field cameras is excluded from the SD9. The customary purpose of this filter is to eliminate 'jaggies' that appear along straight edges in the photograph - a problem that is fundamental to all single-shot imaging sensors".

Footnote(s): The Adobe Digital Video Curriculum Guide PDFs are available from their web site. The reference in this article was to module 2. The thread on dpreview is already generating some heat. Follow the discussion here. Mike Otley has started a thread on Ephotozine. Technical sensor and RAW information drawn inter alia from, Bruce Fraser, 'Real World Camera Raw with Adobe Photoshop CS', Peachpit Press, 2005. pp 235. ISBN: 0-321-27878-X and Tom Ang, 'Advanced Digital Photography, Octopus Publishing, 2003, pp 144. ISBN: 1 84000 6838 and other sources acknowledged in the body text. This article is (c) 2005 John Nuttall and may not be reproduced in any form without prior permission. If linked or quoted, the entire page should be referenced and not a part of it.

Update (1st June 2005): Foveon has published a White Paper by Rudolph J. Guttosch entitled 'Investigation of Color Aliasing of High Spatial Frequencies & Edges for Bayer-Pattern Sensors and Foveon X3 Direct Image Sensors'. Available as a PDF file.

Images on this page (c) 2000-2005 Foveon Inc, Illustrations used by express permission of Foveon Inc.