Sigma SD14: Quick Look at a Quirky Wonder
by Wesley Fink on June 3, 2008 4:20 AM EST- Posted in
- Digital Camera
SD14 Noise over ISO Range
Sigma and Foveon claim the SD14 is 14.06 effective megapixels. It is true that the Foveon image is composed of three dots (red, green, blue) per point, compared to one per point for the Bayer sensor used in other cameras. However, the finished Foveon image is 4.7MPx3 instead of the 10 to 14MP of competing cameras. Of course the extra pixels and the missing colors are interpolated on competing cameras where they are discrete on the SD14, which is why we see the 14.1MP claim. There is no point in taking sides in this argument. You can see and judge for yourself from results.
The output of the SD14 was compared to three other top prosumer cameras in the Olympus E3, Canon 40D, and Nikon D300. The chart below represents one-to-one pixel-level crops of the finished images. Results for the other three cameras were first published in The Digital Sensor: Part 2 and you can go there by clicking the link for more information on how the images were captured.
The Sigma SD14 was a new challenge for this comparison. It is the first SD to offer JPEG capture but most users still consider that capability as somewhat unrefined compared to competing cameras. The SD14 is a RAW camera first that is really designed to be used in conjunction with your computer processing and converting images in Sigma Photo Pro for the PC or MAC. As a result all images were shot as RAW at the same f/4.0 with Incandescent White Balance that was used in competing tests. Images were then converted using Sigma Photo Pro 2.5 in Auto mode and saved as a large 16-bit TIFF. The Tiffs were processed in Photoshop CS3. The 120x200 pixel crops were captured from the TIFF and saved as a Maximum Quality JPEG so it could be displayed on the web.
Our website and most others do not support TIFF images for display, and certainly Foveon RAW is not supported. PNG format would be an option, but PNG files can be 5 times larger than a comparable high quality JPEG with no real improvement in image quality. Therefore the crop and full image were saved from the TIFF in the highest quality option available for JPEG files in the Save for Web option.
Wherever possible the images were captured using a 50mm f/1.4 normal lens. This represents an equivalent 35mm focal length of 75mm for the Nikon D60, and 80mm for the Canon 40D. A 50mm on the 4/3 cameras would have an equivalent 100mm field-of-view so images were captured using the 35mm Olympus Macro lens, which is equivalent to 70mm. This lens was chosen because it is critically sharp wide-open and is in its best resolution range at the standard f/4.0 capture aperture.
The SD14 images were captured with a Sigma 50mm f2.8 Macro lens I purchased many years ago with an SA-300N camera. It was and is a very good Macro lens and it worked perfectly on the SD14. Keep in mind the Sigma SD14 has a lens multiplier of 1.7x so it is equivalent to an 85mm lens on a 35mm camera. That accounts for a small portion of the field-of-view variations in the equal pixel crops. However, most of the difference is a result of the differing pixel size of the finished image
All images were captured at the same f/4.0 aperture using a tripod in the same location. Focus was manual and the camera program selected the shutter speed. Lighting was a single 100W Tungsten bulb high right, and all cameras were all set to the Tungsten (Incandescent) preset.
| Sensor
Performance JPEG Comparison over ISO Range |
||||
| ISO | Sigma
SD14 14.06MP (4.7MP x 3) |
Canon
40D 10.1MP |
Olympus
E3 10.1MP |
Nikon
D300 12.3MP |
| 50 |  |
|||
| 100 |  |
 |
 |
 |
| 200 |  |
 |
 |
 |
| 400 |  |
 |
 |
 |
| 800 |  |
 |
 |
 |
| 1600 |  |
 |
 |
 |
| 3200 |  |
 |
 |
|
| 6400 |  |
|||
Click on any of the above image crops for the full image. Note: Full size images are between 2.0MB and 5.8MB! |
||||
The Foveon is widely considered a low ISO sensor. In the second generation Foveon, most consider best performance is ISO 50 to ISO 200. There is no doubt in looking at the original RAW images that noise starts to become apparent at ISO 400 and it becomes pronounced at ISO 800. Looking at an ISO 1600, the noise in the RAW image makes it all but unusable. Our surprise, however, was that the latest version of Sigma Photo Pro 2.5 does a pretty decent job of minimizing noise as the images are saved to TIFF. 400 is certainly usable , as is 800 for all but large prints. ISO 1600 should be reserved for those situations when nothing else will produce an image.
The sweet spot for the Foveon is definitely ISO 50 to 200. Image quality is superb at those speeds, but 400 and 800 are also usable. The Foveon is not sensitive like some of the best of the current CMOS sensors. It is nowhere near the ISO range of those new sensors. However, it does reach lower, to ISO 50 in the latest firmware than its competitors. You will find ISO 50 as an extension on the Canon 5D and on other full-frame sensors, but it is a rare option on crop sensors.
Facebook
Twitter
RSS
40 Comments
View All Comments
chibimike - Tuesday, June 3, 2008 - link
Layer-per-primary is not the future.It is a great idea, but passing through the first two layers to get to the red layer eats too much of the light. To make up for it, they make the red layer supersensitive, which then leads to clipping or extremely complex attempts at color correction. Unless there is some breakthrough material to make the sensor out of, it will always be noisy.
aeternitas - Tuesday, June 3, 2008 - link
I see your point, and its a good one.Except useing this method of three layer technology and pretending thats the only way is pretty close minded and short sighted.
Their are better methods than what Foveon has, but layer-per-primary as a base WILL be the future of digital photography. Its a technology in its infancy, and hinestly, do you see any other method that would even come close to haveing the same future potential? If you mention fuji ill shoot you.
Justin Case - Wednesday, June 4, 2008 - link
Potential is irrelevant when the current technology has aready hit the "lens wall".Bayer pattern sensors are already at the limit of lens resolution, and can still be pushed forward. Unless Foveon can suddenly increase the resolution of their sensors by 300%, fix all the color problems, and Sigma releases lenses that can beat the top Nikon and Canon models, they simply don't offer any advantage (note that there are sharper lenses than Canon's or Nikon's - Leica, Hasselblad, Voigtlander, Zeiss - but they're used by less than 0.1% of people).
If anyone is interested in improving the quality of Bayer sensors, the way to go is tetrachromacy (Sony is experimenting with that), but that makes processing more complex and offers only a very slight little advantage, so I don't see that happening, either.
Current Bayer sensors have hit the resolution limit for lenses, and improvements to sensitivity and chroma range are more important than resolution. Since those are precisely the points where Foveon is further behind, its future isn't particularly bright. Some people will continue to buy them, because the concept "seems to make sense", but in the real world they're simply worse in every sense. Even if done "perfectly" (with materials that don't even exist yet), they'd be more expensive and the advantage wouldn't be noticeable without a major upgrade to lens quality.
chibimike - Wednesday, June 4, 2008 - link
No not fuji, great for pocket cams, not so hot for SLRs.I think one of two things. Either a new material with much lower noise properties. Or if costs keep going down, we could potentially see someone do a creative take on the three sensor design like they do in high end video cameras.
aeternitas - Tuesday, June 3, 2008 - link
PS.There is this new and great technology called PNG. No one expects you to display tiff or raw files. I'm not sure why a paragraph was put into explaining that to everyone.
Wesley Fink - Tuesday, June 3, 2008 - link
We use PNG all the time in our articles and reviews and it is a desirable lossless format like TIFF. However it is not really very suitable for photographic images. To quote from Wikipedia:"JPEG can produce a smaller file than PNG for photographic (and photo-like) images, since JPEG uses a lossy encoding method specifically designed for photographic image data. Using PNG instead of a high-quality JPEG for such images would result in a large increase in filesize (often 5–10 times) with negligible gain in quality."
For photographs PNG is no better than JPEG, but the files are MUCH larger. PNG is best with tables and line drawings. We need a new web format as neither TIFF or PNG are suitable and current JPEG is just 8-bit - perhaps an update of JPEG.
Justin Case - Wednesday, June 4, 2008 - link
PNG uses prediction and error coding, followed by entropy encoding. It's pretty much the best lossless algorithm you can get for photos.For line drawings and tables, GIF (LZ compression) will frequently produce smaller files than PNG, especially if they have regular dithering patterns.
If you think there's no difference between JPEG (even at the maximum quality allowed by Photoshop) and PNG, you need to have your eyes checked. Even though Wikipedia is hardly an authoritative source, the article you quoted says "a negligible gain in quality". Not "no gain". You're the one who extrapolated that to "PNG is no better than JPEG".
The meaning of "negligible" depends on the situation. Yes, the PNG files are significantly bigger. And yes, if you use a crappy camera, you'll get artifacts from the lens and sensor anyway. But when you're comparing high-end cameras and sensors, it definitely pays to use a lossless format (generated from raw, of course, not from an in-camera JPEG). Even better, post a couple of raw files and let people who are interested download and compare them using appropriate software.
BTW, there already is an "update of JPEG". It's called JPEG-2000 and gets about 40% higher compression for the same (lossy) image quality, and even supports lossless compression. If you can convince browser makers (namely Microsoft) to add support for it to their software, we'd be thankful. Until then, I doubt it'll catch on.
I have no idea why you need a higher-than-8-bit format "for the web", considering that about half the LCDs out there are actually 6-bit panels and that 99.9% of graphics cards are limited to 8-bpc. We definitely don't need browsers to become more bloated just so they can handle HDR or RPF images natively (which 99% of people have no use for in web pages).
For distribution and webpage "decoration", 8-bit JPEGs are fine. The question is whether they're fine for an article comparing minute differences between high-end digital sensors. And the answer to that is almost certainly no (just as MP3 files aren't really the right choice to evaluate high-end audio equipment).
Deadtrees - Thursday, June 5, 2008 - link
Last time I checked, JPEG2000 is pretty much dead and MS is trying to setup a new JPEG2000 like picture file system along with many other major companies. I forgot the details but it was open architecture and it was better than JPEG 2000.Justin Case - Friday, June 6, 2008 - link
JPEG-2000 is only "dead" due to lack of native support in Windows and web browsers. It's a significant improvement on JPEG.Microsoft's format (which has gone through three names, I think the current one is "MS HD Photo") is a DRM'd format which, in terms of quality vs. compression, is slightly inferior to JPEG-2000 and can only be manipulated through Microsoft's own APIs (very much not an "open architecture"). It does have support for several useful features (like floating-point HDR) but I really don't see those being relevant for a distribution format; if you want HDR data, you probably don't want lossy compression, and vice-versa.
A unified "raw" format for all camera manufacturers, on the other hand, would be nice, but considering the different sensor layouts, different bit depths, different processing required, etc., you would have a single file extension but effectively N different sub-formats.
aeternitas - Tuesday, June 3, 2008 - link
I understand the differance between the two.In context, when mentioning all those formats, and not mentioning png, its kinda off. Also, PNG with such tiny crops is negligible.
Aside from that, a contrustive suggestion would be, if youre going to do side by side images, have a single croped image, and several mouse-overable links that change the image dynamicaly. It gives a much much better result when trying to see subtile differences between cameras.
IMO that would one-up anandtech camera reviews compaired tomost online.