Brian's Brain

EDN Senior Technical Editor Brian Dipert exposes, analyzes and
opines on diverse topics in technology.

Profile

About Me

RSS Feed

Add this blog to your RSS newsreader!

By Category

Consumer Electronics Design Articles

Blog

Brian's Brain

« Apple's iPhone: More Initial H... | Main | Bunnie and Chumby In China... »

Friday, July 13, 2007

Sigma SD14 Dissatisfaction: Finally Believe Me, Foveon Fanboys?

I've written about image sensor supplier Foveon and its (currently only) camera OEM partner, Sigma, on multiple occasions: within two print feature articles ('Imaging Beyond Pixels: Enhancing The Elementary Picture' and one of its online addendums, along with the earlier 'Seeking clarity: Image sensors peer into a blurry future'), as well as in a number of blog posts. My consistent message has been that Foveon's Mpixel claims can't be taken seriously; that simply by embedding discrete red, green and blue spectrum-optimized photodiodes within each photosite (pixel), Foveon can't claim that they've tripled the sensor's resolution. This stance is especially valid when the alternative Bayer pattern sensor, whose proportion of red, green and blue filters synergizes with the human visual system's sensitivity to those portions of the visible spectrum, is as effective (in conjunction with interpolation, to create the missing remaining spectrum information for each pixel) as it's been proven to be over more than 30 years' worth of development and commercialization.

There's nothing like hard data to back up a theory-determined gut feel. Following up in their earlier initial hands-on impressions, Popular Photography has undertaken an exhaustive test of Sigma's latest SD14 camera (which I reported on back at PMA) and the Foveon '14.1 Mpixel' (4.7 million photosite) X3 sensor inside it, benchmarked against the Nikon D80 and its 10 Mpixel sensor. Both reports are excellent and I encourage you to read them in their entirety; to whet your appetite, I'll share a few choice quotes from the latter writeup's overview page below:

Black-and-white test targets for measuring resolution don't show as much detail as Foveon's 14.1MP count implies. Analysis of the IT-10 black-and-white resolution target we use in the Pop Photo Lab finds the Sigma SD14 on par with a good 8-9MP camera (in RAW mode), but not in the same class as 10MP models such as the Nikon D80.
Using a color spiral target that Foveon sent us for additional tests, we found the SD14 captures slightly more detail in red-and-blue areas than the Nikon D80 -- but not in the target's black-and-white or dual-color areas that contain green. So, in real-world shooting -- where it's hard to find finely detailed red-and-blue subjects or scenes that have little green or b&w detail -- the D80 has the sharpness advantage and outperforms the SD14 at high ISOs.
Foveon and Sigma are right in saying that color moiré in very fine detail was a problem in early digital cameras using Bayer pattern sensors. However, most manufacturers now use low-pass filters and image processing to minimize or eliminate moiré. The loss in resolution goes unnoticed because of higher pixel counts.
RAW data from the X3 sensor still seems to require intensive (not minor) postprocessing to convert color information into the final image. And more processing translates into a slower burst rate on the SD14 compared with other DSLRs. In-camera JPEG images from the Sigma SD14 also show less color accuracy and detail than RAW files processed by a computer. (Editor note: sorry, Analog Devices, better luck next time....)

This quote's my favourite, and from the camera review:

We discovered, though, that the 14MP Super Hi JPEG format is merely an interpolated version of the 4.7MP Hi Quality JPEG, taking up much more storage space without offering a significant image-quality or print-size advantage.

Granted, the resolution captured by a camera is a complex combination of multiple factors, not just determined by the image sensor specifications, as imaging guru Mark Schubin has explained on numerous occasions (most recently in an outstanding writeup in Videography magazine). It's possible, for example, that the SD14's optics are holding back the Foveon sensor potential....but given that Sigma's first and foremost a lens manufacturer, I frankly find this particular hypothesis to be quite doubtful.

You've previously read my thoughts on Foveon. Now you've got the supportive statistics. I rest my case. Comments, folks?

p.s....speaking of sensors, and the gadgets containing them, check out the thought-provoking discussion 'Digital Camera vs. Camera Phone' that recently appeared on Slashdot.

Followup: Nothing beats a price cut when striving to stimulate sales, I guess, although in comparison to deals like this one, the SD14 still seems way overpriced to me.

Reader Comments

(Add your comments)

at 7/13/2007 10:07:38 AM, Curious said:
So ... if more samples per pixel location do not improve image quality, why is it that the highest quality video cameras use three sensors and a prism?

at 7/13/2007 10:49:38 AM, Brian Dipert said:
Dear Curious, I never said that three samples per pixel doesn't improve image quality, but there is a tradeoff....the Foveon sensor is effectively only 4.7 Mpixels in resolution, whereas 10 Mpixel Bayer pattern-based CCD and CMOS sensors are now prevalent. And to answer your question, videocams use a three-sensor prism arrangement both to improve image quality and to improve low-light sensitivity. However, this is (of course) a much more expensive, bulky and power-hungry approach compared to the single-sensor Bayer or derivative technique

at 7/13/2007 6:09:49 PM, Josh said:
Bayer cameras use over half of there "mega-pixels" for resolving power...a ten mega-pixel bayer camera isn't really a ten mega-pixel camera, just like the SD-14 isnt a true 14 mega-pixel camera.

at 7/15/2007 1:24:56 AM, Laurence Matson said:
You state: ". . . that simply by embedding discrete red, green and blue spectrum-optimized photodiodes within each photosite (pixel), Foveon can''''t claim that they''''ve tripled the sensor''''s resolution." Could you please cite one piece of literature from Foveon claiming that they have "tripled the sensor''''s resolution."? Remember, resolution is the operative word in the above. And while you are at it, please explain how spatial locations are related to resolution. (That might be easier.) Finally, since you are a big reporter at PMA, you obviously missed the amazing video presentation by Foveon at PMA 2005!

at 7/15/2007 3:06:03 PM, Brian Dipert said:
Dear Laurence Matson, peruse (for example) the Wikipedia definition of 'pixel'. Both in the image capture and image display realms, common and longstanding industry convention refers a pixel as 'the smallest complete sample of an image'. The word 'complete' is key here, as is another portion of the Wikipedia essay; 'in color systems, each pixel has typically three or four dimensions of variability such as red, green, and blue, or cyan, magenta, yellow, and black.' Now look at the X3 Technology overview writeups on Foveon's website. Foveon has long attempted to redefine 'pixel' to mean an image capture element, convenient because in doing so it enables them to effectively triple their sensors' claimed pixel counts (i.e. their 'resolution', again referencing common and longstanding industry convention reference of the term). I'll directly quote one example of Foveon's attempt to rewrite history for you (reference: www.foveon.com/article.php?a=68): 'A pixel on the image sensor of a digital camera is a light absorbing element (photodetector) that converts light (photons) into electrons. A pixel is also referred to as a pixel sensor when there is a need to distinguish the pixel from its location. A pixel location is the X,Y coordinate on the two-dimensional grid of an image sensor at which the pixel is located.' This is absolute rubbish. We don't refer to the red, green and blue sub-elements within each LCD pixel as 'pixels', we refer to them as sub-pixels. Similarly, it makes no sense to refer to the red-, green- and blue-tuned photodiodes within each Foveon sensor pixel as pixels in and of themselves....call them subpixels, if you wish, but NOT pixels. I've never disputed the fact that Foveon's approach MAY provide more accurate color than a Bayer patterned-sensor of equivalent pixel resolution. But is Foveon's latest X3 sensor '14.1 Mpixel' (reference: www.foveon.com/article.php?a=222)? No, I've consistently said. And Popular Photography has just confirmed my stance. End of story.

at 7/16/2007 2:17:08 PM, Cristian Oltean said:
I would not use Wikipedia as a technical reference. And the definition ''''the smallest complete sample of an image'''' has no meaning. It may get one in a different form like ''''the smallest complete independent sample of an image''''. Independent is the key here. This is why interpolation is not accepted as a method to increase the real resolution. Because pixels would be no longer independent. But then not even the professional cameras deliver fully independent pixels there is always some blurring. And now the big question: how much cross-correlation is acceptable between adjacent pixels?

at 7/16/2007 5:31:09 PM, Brian Dipert said:
Dear Christian, I'll requote from Popular Photography's writeup, which echoes comments I've made several times in the past: "Foveon and Sigma are right in saying that color moiré in very fine detail was a problem in early digital cameras using Bayer pattern sensors. However, most manufacturers now use low-pass filters and image processing to minimize or eliminate moiré. The loss in resolution goes unnoticed because of higher pixel counts." Now granted, if you do a wall-sized enlargement of a tiny portion of the original image, you might still see a per-pixel color accuracy difference with a Foveon sensor-based camera versus a Bayer pattern alternative....but then again, I'd argue you'd be much less likely to do such a thing with a Foveon-based camera because of its inherently lower pixel count! And I agree with your fundamental distaste of interpolation, by the way....I've never been a fan of 'digital zoom' (as my writeups have, I think, made clear).

at 7/18/2007 8:55:46 AM, Michael J. McNamara said:
Brian: As the author of the report (s) in Popular Photography and Imaging, I''m of course pleased to find someone who agrees with my findings. I''d only add that I think both Foveon and Sigma did themselves a dis-service by hyping the pixel counts as they have for both generations of X3 sensors. The fact is that for a 4.7MP layered-sensor the current one in the SD14 resolves at the equivalent level of an 8-9MP Bayer-pattern sensor, and it records some color details better (but not all). Why couldn''t Foveon just promote that fact and not try to "rewrite history"? By the way, Foveon attempts to rewrite history again on its web site with the following quote: "For over 100 years, color film has traditionally been held as the gold standard for photography. It produces rich, warm tones and incredible color detail that consumers around the world have become accustomed to. Film has achieved this by using three layers of emulsion to capture full color at every point in the image." WIthout even checking my Focal Encyclopedia of Photography I can tell you that color film hasn''t been around for 100 years. But that encyclopedia gives credit to Kodak in 1935 for the intro of Kodachrome as the first commercially viable color film. Prior to that, color photography and printing relied on screening and filter techniques that recorded separate color data on panchromatic film and required precise registration in both capture and printing. That is actually a process that it closer to mimicking the Bayer Pattern! My test of the SD14 also pointed out the failings of this camera and sensor at higher ISO''s (above 400) just as my tests on the original SD9 pointed out its failings in exposure latitude compared to typical CCD and CMOS sensors. For the price of the camera, these performance problems made it non-competitive with other cameras such as the Nikon D200 or Canon EOS 30D.

at 7/18/2007 2:05:02 PM, Rick said:
In the Analog Devices press release link ("sorry Analog Devices"), Sigma COO Yamaki states the "SD14 tells the ''whole truth'' about an image, by capturing the real color of every pixel." That can be interpreted either the pixel captures the real color (which would take the 3 subpixel elements) and would be a 4.7 MP camera, or each element captures the real color. The second interpretation would be inaccurate in that only one portion of the real color is caputured (red, green, or blue) and not the true color. Looking at the Sigma website, they would seem to pull back from the pixel count argument by saying "it''s not all about the pixel count" (philosopy section). But then they go on and say "Its high image quality comes from 14 megapixels" so maybe they are hung up on pixels. Sounds like they are trying to play the megapixel marketing game that people seem to buying. Pixels shouldn''t matter as long as there are enough of them to give good resolution for the task and the image looks good.

at 7/26/2007 9:08:07 PM, John Clifford said:
Using ''''pixel'''' to refer to resolution is invalid, because a pixel is the final ''''picture element'''' that is the result of interpolation from the raw Bayer sensor data. All Bayer images are interpolated, which is a fancy way of saying ''''the result of guesswork'''' and it is a credit to modern software engineering that interpolation routines guess as well as they do. Therefore, an interpolated 10 MP JPEG image from a Bayer sensor no more reflects 10 MP of true resolution than a 14 MP Foveon JPEG image from the SD14 reflects 14 MP of true resolution. Interestingly enough, though, a 4.2 MP image from a Foveon sensor has about the same amount of resolution as a 10 MP image from a Bayer sensor... because the smallest item that can be resolved by a Bayer sensor spans a circle that has to hit at least four adjacent ''''pixels'''' (photosites) on a Bayer sensor, while it only needs to span a circle that hits a single ''''pixel'''' (photosite) on a Foveon sensor. The best term to use to discuss image sensor resolution is ''''photosite'''' which is the smallest sensor area that can detect detail

at 7/26/2007 9:15:57 PM, John Clifford said:
The best term to use to discuss image sensor resolution is 'photosite' which is the smallest sensor area that can detect detail; on a Bayer sensor it is comprised of four adjacent photodetectors (incorrectly referred to as sensor pixels) arranged in a 2x2 horizontal matrix, the photosite area being a circle that spans the 2x2 matrix. On a Foveon sensor, a photosite is comprised of three vertically stacked photodetectors with the sensor area of a single photodetector. I agree that Sigma's resolution claims (14 MP) are misleading, but so are the resolution claims put out by the various camera manufacturers using Bayer sensors. In the end, what counts is results, and the Sigma SD14 equals and often exceeds its 10 MP Bayer competition in real-world (not black and white resolution target) resolution tests. I eagerly await the next generation of Foveon sensors, and wait for the day that a full-frame 8 MP Foveon dSLR sensor is introduced; it will blow away everything else in the dSLR market including the 16 MP Canon MkII.

at 7/27/2007 12:13:42 AM, Laurence Matson said:
Brian, in your long comment here: at 7/15/2007 3:06:03 PM, Brian Dipert said: You did not answer any of the questions asked. You don''t have to humor me with lengthy discussions on pixels with all kinds of cites to Foveon and Wikipedia. I probably know all of that at least as well as you if not better. But why not address the question, which was, again: at 7/15/2007 1:24:56 AM, Laurence Matson said: You state: ". . . that simply by embedding discrete red, green and blue spectrum-optimized photodiodes within each photosite (pixel), Foveon can''t claim that they''ve tripled the sensor''s resolution." Could you please cite one piece of literature from Foveon claiming that they have "tripled the sensor''s resolution."? Remember, resolution is the operative word in the above. And while you are at it, please explain how spatial locations are related to resolution. (That might be easier.) Finally, since you are a big reporter at PMA, you obviously missed the amazing video presentation by Foveon at PMA 2005! Specifically, what do pixels have to do with resolution? Where does Foveon state that they tripled the resolution? And while you are at, could you please explain why there might indeed be some advantages to this approach in the form of luminance and chroma resolution? If you were truly curious about pixels and definitions, you would delve into the source material from the defining bodies and not run around Wikipedia. Also, Dick Lyon''s lecture on pixels given at Google is an excellent starting point for the open minded on coming to terms with this defining pixel. Since I suspect that you make your micro-bread more by taking a stance and sticking to it than really digging into the subject and perhaps coming to a different conclusion from that at the starting point, I guess that you will just stand behind to CFA barriers in your mind and keep those clicks coming. But I am willing to be surprised.

at 7/27/2007 1:33:19 AM, Robert said:
Wikipedia isn''t always reliable, but in this case their definition of pixel is consistent with everyone elses definition, except Sigma and Foveon of course. Pick any engineering textbook or Google for it. Laurence, you''ve been thru this debate in various other forums and you still refuse to accept the correct definition. Why? The sensor has 4.7 million pixels. Calling it 14.1 million is tripling it. I''m not sure how much clearer it can get. Sigma originally called the SD-9 a 3.4 megapixel camera, but after low sales they tripled it it to 10.2 megapixels with the SD-10. The sensor didn''t change (other than microlenses). The difference was nothing but marketing. There is no technical reason for that (ask any engineer). I also must point out that Bayer does NOT guess at all. It accurately calculates the missing information and it can accurately resolve finer detail than a 2x2 grid. Foveon also calculates RGB from its three samples (which aren''t pure RGB), and its a more complex calculation than Bayer. Why isn''t that called ''guessing'' ?

at 7/27/2007 2:03:44 AM, Guido FORRIER said:
The sd14 foveon has 4,7 million pixels for all the colors . A bayer 10mb has 5 million for green , 2,5million for blue and red .

at 7/27/2007 6:41:24 AM, Mike said:
I find this a rather bold article for someone who (apparently) has not even held the SD14 camera and is just picking up some random pieces off the internet. Sigma would have gotten blasted on "pixel" count no matter which way they cut it. It doesn't matter how you define "pixel", the fact of the matter is that the Foveon technology is different so something isn't going to match. Mr. Dipert chose to use the Wikipedia definition of pixel, so let's stick with that. By that definition of pixel, Nikon shouldn't be able to claim 10 megapixels for a D80 either because there isn't a single photosite on the sensor that is complete! See where I'm going with this? Had Sigma used 4.7 MP, something would have been lost in the translation. They use 14 MP which is also not a good/accurate term but had they used 14 megasensor, 14 megadiode, or even 4.7 megapixels x 3 layers, people would have no idea what that means without having to read an article. In my eyes, Sigma/Foveon were damned if they do, damned if they don't. And your statement about Bayer cameras being refined over the years is true, but certainly not a valid negative for Foveon. Bayer has been around for years too with video cameras but it is a well known fact that 3 chip video cameras perform better. Replicating the DEFICIENCIES of the human eye is not a good thing! The goal of digital photography should be to produce the most accurate result possible of the original scene and (good) full color capture will always perform better than Bayer for that purpose. Where the equivalency is and whether a 4.7 MP full color capture sensor is a good match for an 8, 9, or 10 MP Bayer sensor, that gets a bit more complicated.

at 7/27/2007 7:14:46 AM, Sean said:
As a bayer user, i have been very intrigued by the potential of the Foveon technology. Bayer has very visible shortcomings in what I would call "hard color" transitions (like red to white). You can see a noticeable drop in image detail at these transitions. On the other hand, these transitions are not all that common and are not easy to see if you are not looking for them. There is alot of mis-information spread around in these discusions (my favorite lie is that bayer cameras use 4 photosites to make 1 pixel, giving you 25% resolution). The truth is that the SD14 is a true 4.7mp camera, while a 4.7mp Bayer camera would not provide the same amount of detail. The problem Foveon has is that they are not competing against 4.7mp cameras, they are competing against 8 and 12mp cameras.

at 7/27/2007 7:49:20 AM, Brian Dipert said:
Dear Robert, I wasn't aware of the resolution spec inflation (with no change in underlying sensor) that paced the SD-9 to SD-10 transition. Thanks for the historical data; that's a classic!

at 7/27/2007 8:07:32 AM, Ed_S said:
Dear Brian, As this is my first exposure to your work I have limited basis for evaluating its scientific or photographic worth. I must politely comment that as I read this article and the various comments/responses, I'm a bit put off by your apparent "agenda". The very tone of the title is offensive IMHO to any satisfied user of Sigma DSLRs and if you will "poisons the well". If you have to say "fanboy" you are not objective. You make several worthy observations to include the point that 4.7M x 3 may not be 14M effective megapixels. Many knowledgeable as to Sigma/Foveon technology would agree, but would counter as has at least one reader that 14M of bayer really doesn't produce 14M of resolution. Coupled with that, your arguments rely (apparently in this article) primarily on secondhand theory and opinion to include the less-than-unbiased evaluation in Popular Photography. As a user of Bayer, Foveon, and even 3CCD equipment, I welcome a scientific and unbiased discussion which sets out to present fact as best current science knows it. I'm afraid that your article at best sees the world through a Bayer sensor. Or so it seems to me.

at 7/27/2007 11:45:29 AM, Peter Spader said:
The pixel count problem arose due to the fact that the conventional way of sizing sensors does not take into account the vertical stacking (VFA) Foveon used. Initially Foveon tried to indicate the difference my simply adding X3 (time 3), but people started classifying the SD9 as just a 3.4mp camera, so Foveon did the multiplying and called the SD10 a 10.2mp camera and the SD14 a 14mp camera. Given the fact that a case can be made that the CFA (aka Bayer) is not truly an Xmp camera, we really do need a common designation everyone can agree to. This will be hard enough from a technical point of view, and once the marketeers get involved almost impossible The more general problem is trying to come up with a fair way of comparing the Foveon/Sigma cameras with the CFA (aka Bayer) cameras given just how different these two approaches are. Take resolution (only one element of IQ), for example. The conventional B&W resolution charts do not work well, and the various attempts to come up with alternatives have not yet gotten to the point where everyone agrees. Furthermore there are other factors in IQ resolution does not address. Which leaves us with the “subjective” reaction to the images themselves. I personally find the Foveon images captivating in a way the CFA images are not and have taken part in some of the discussions trying to find out why. There are a number of factors that have been suggested, some of which are not just “subjective,” but again, we are not yet at a point where everyone agrees. At the same time I have never suggested the best CFA images are not very good indeed. I am less interested in the “which is best” debates since they often raise the number of variables to the point where the whole debate becomes a set of crossed monologues. To the degree they explore new ways of making comparisons they have some value, however. BTW, ad hominem does not help much either. I am a fan of IQ, but at the age of 68 I am no longer a boy, and if Fan Boy is meant to suggest I am uncritical, that too is inaccurate. Given the hope that we are all interested in the furthering of “the catching of light,” as some called photography, lets keep the debates focused, please.

at 7/27/2007 11:51:43 AM, John Clifford said:
How can a Bayer CFA sensor calculate the color on the area of a single photosite, when that single photosite has a color filter that blocks two of the three primary colors? By looking at adjacent photosite data, and GUESSING that, based upon a very complicated algorithm, given a certain data value at this and adjacent photosites an assumption can be made that the color at this particular photosite is "X". Any change in color that is too small to hit at least three adjacent photosites is subject to miscalculation, i.e., a bad guess. The Foveon approach, on the other hand, has three separate color-sensitive photodetectors in the same area (because they're vertically stacked). Therefore, changes in color that hit a single photosite are detectable... unlike with a Bayer sensor. Want proof? Take a 50mm M42 lens with an EOS adapter, put it on a Canon 5D, and take a picture. Put the same lens, with an SA adapter, on an SD10. A 2268x1512 center crop out of the 5D's image will be identical to the image from the SD10 in terms of image coverage because the photodetector (pixel to you non-technical types) size is almost identical. You will notice that fine detail which cannot be seen on the 5D image is clearly visible on the SD10 image. Why? Because the 5D's image processing algorithm guessed wrong, and because the SD10's image processing algorithm didn't have to guess.

at 7/27/2007 11:55:30 AM, John Clifford said:
BTW, if you really don't think Bayer CFA sensors guess, then I have just one word for you... MOIRE.

at 7/27/2007 12:42:23 PM, Walter said:
I''m not convinced that Sigma is right to use these pixel counts - but they aren''t the ones who started mis-leading the public - and their action is certainly as fair as those of the Bayer crowd. In *any* other application, like an LCD screen for example, a pixel is made up of red, green, and blue elements. You don''t call your 1024x768 LCD panel 3072x768 or 1024X768x3 - though any current camera company would - the first is what Bayer cameras do, and the second what Sigma is doing. That''s right. Bayer MegaPixels are just wrong. 2/3rds of the information in every output pixels is an interpolated guess. This is where Moire comes in, and the reason you have to sharpen Bayer images, either in camera or in development - your "native" resolution is substantially interpolated to begin with. The fact that sharpness varies in a natural, optical fashion creates the "3D" look of the sigma, which Bayers can only achieve by downsizing. So Sigma gave an honest 3.4 megapixels to the SD9 - and the world ignored it because the Bayer crowd was providing a dishonest 6-8 MP and the same time. (which the SD9 generally out resolved.) In choosing to call the SD10 "10 megapixel," they only committed the same deception that all other digital cameras have except for the few monochrome models out there. I don''t generally consider myself a "Sigma Fanboy," but the answer to your question is no, I''m not at all convinced. The Sigma rating is at least as accurate as that of competing Bayer Cameras, and test pictures from a dozen sources show it matching resolution with Canon''s magnificent 5D on even terms. (Note, pictures, in the real world, not B&W test charts that favor the asymmetric color resolution of a Bayer sensor.) Walter

at 7/27/2007 3:35:07 PM, Kendall said:
1) Color processing is not done in-camera (unless you shoot JPEG). And JPEG images write out quicker to a CF card than RAW (because there is less data). You need a new theory on processing time for color conversion. 2) No red & blue subjects in the real world? What about just about anything with fine detail against the sky? Or red cars? Or Fall leaves against a blue sky? Or Red boats in the water? It doesn't take much to think of many things that come across being helped by red & blue resolution enhancements. But even green is helped, which you can see when looking at foliage details like distant leaves or blades of grass. 3) If it's really a 4.7 MP camera, how is it that it has the same evident detail as a D200, or even a 5D in some cases when enlarged? The problem is you are using a fixed number from a bayer camera that represents a variable level of output quality, to try and compare against a fixed number on the SD-14 that represents a fixed amount of detail captured regardless of color content. Anyone can take a look at any one of numerous full-size SD-14 images (such as on www.pbase.com) and see there is more data there than a 4.7MP bayer camera - by stating this you only hurt your own credibility.

at 7/27/2007 3:42:03 PM, Kendall said:
Robert: What the Foveon chip does is not guessing, because for any given location on the image it has the full spectrum of data availaible. If anything the process by which it determines color could be called "untangling" because there is some overlap between the three sensors - but because there is no break in data there is no "guessing" to be done. <p> As for pixels, you are confused because camera sensors use "photosites" to capture data that are converted into the "pixels" you see on screen.

at 7/27/2007 4:45:57 PM, Alf Beharie said:
Brian, the problem is you are trying to think of Foveon sensor pixel count in a Bayer sensor pixel count mindset, whilst not understanding how Bayer pixelcounts are wildly over exagerated for no other reason that to help sell more cameras...Here is a quote direct from reference.com, the online encyclopedia, which should help you understand that Foveon pixel counts are just as valid and accurate as those used for Bayer sensors: "In contrast to conventional image sensors, the Foveon X3 sensor uses three layers of sensor elements, so that it detects red, green, and blue intensity at each array location. This structure eliminates the need for de-mosaicing and eliminates the associated image artifacts, such as color blurring around sharp edges. Citing the precedent established by mosaic sensors, Foveon counts each single-color sensor element as a pixel, even though the native output file size has only one pixel per three camera pixels. With this method of counting, an N-megapixel Foveon X3 sensor therefore captures the same amount of information as an N-megapixel Bayer-mosaic sensor, though it packs the information into fewer image pixels, without any interpolation."

at 7/27/2007 6:14:06 PM, Oliver Nielsen said:
How the Foveon sensor compares, when shooting spirals and charts in a lab is of little interest to me, as I pride myself in LOOKING with my EYES on PICTURES of REAL OBJECTS, and judge the camera on that.

at 7/27/2007 6:21:51 PM, Oliver Nielsen said:
What's good is, all this anti-hype against Foveon is actually good advertising pro-Foveon and Sigma. I became a Sigma SD14 user last week, partly because of reading all these discussions and then playing around with some of Carl Rytterfalks RAW files. I could have purchased a Canon 30D instead but really don't like Canon's porky, flat, two-dimensional take on skin tones. The Sigma's images are not perfect, but they do have that depth I missed since selling my Pentax 67 on eBay. You want Canon? Get a Canon. You want Sigma? Get a Sigma. Nikon? Go ahead... In the end, I feel confident my Sigma will give me the results I need. Could those results be achieved with a Bayer based camera? Yes, but I feel Foveon gets me there faster and better. Photography is about personal expression. Then for gods sake let the tool chosen be a personal choice of the individual. Being a photographer is not organized religion.

at 7/27/2007 6:33:11 PM, Brian Dipert said:
Dear Oliver Nielsen, I'm still holding onto my prized P67 and its lens suite and other accessories....and the film for it (from a Roseville, CA photographer who 'went digital') is sitting in the freezer awaiting my attention!

at 7/27/2007 7:28:07 PM, Robert said:
There is no change in pixel count by having multiple layers per pixel. None whatsoever. Bayer does not guess pixels, it calculates them, as does Foveon, but with different formulas. Moire is not caused by guessing, it is from alias artifacts, which on a Bayer sensor are colorized (and rare). Sigma cameras have no anti-alias filter so there are a lot more artifacts (but not colorized), which many people confuse for more detail. Sigma also sharpens the image more than other cameras. Not a fair comparison, but it does have a unique look. Comparing SD10 to 5D and cutting out almost all of the 5D image is absurd! What kind of crazy biased test is that? This page is linked at the Sigma forum at dpreview and they are having a field day with it, which is why there are a lot of comments suddenly.

at 7/27/2007 11:35:45 PM, John Clifford said:
Robert, you do not, and cannot, explain HOW a Bayer sensor-equipped camera calculates the RGB value of a final pixel (picture element) based upon a monochromatic image sensor pixel (photosite). The fact is, it interpolates, i.e., GUESSES. Now, the guess is a very educated guess, based upon the values of adjacent photosites, but it is STILL a guess, and that fact that the guesses are right far more often than wrong is a tribute to smart software developers. When the camera gets it wrong, then you get wrong colors, or in the extreme you get MOIRE, a.k.a. INCORRECT GUESSES, a.k.a. alias artifacts. The Foveon sensor doesn't have to guess since it can sample each primary color at the same photosite, thus detail only becomes ambiguous when it spans less than a photosite. The same is true of a Bayer sensor, but the photosite area is greater than individual photodetector (pixel to the ignorant) area, and thus for a given photosite density the Foveon will resolve greater detail. The cropped 5D to SD10 comparison IS a fair comparison AT THE LEVEL OF THE INDIVIDUAL PHOTOSITE and shows how you get more resolution for the same image area, given the same photosite size, on a Foveon sensor versus a Bayer sensor... no one is saying that an SD10 has more photosites, but that the resolution on a per-photosite basis is noticeably better. And finally, the default Sigma SDx image format is RAW (X3F), and no sharpening is applied.

at 7/28/2007 2:38:21 AM, Alf Beharie said:
The man who conducted that SD14 vs Canon EOS 5D comparison you refer to did not cut out any of the 5D image to match the crop of the SD14. I think you will find that he actually used a wider lens on the SD14 so that the FOV's were virtually indentical (SD14 + a 50mm lens x 1.7 crop=85mm vs 5D + 85mm lens=85mm)and then he upsized the resulting SD14 shot to match the size of the 5D shot which gave the 5D something of an advantage yet it still could not capture fine detail in greens as well as the SD14. If you think the methods employed in some way biased the result in favour of the SD14 personally I think you will find its the opposite but if you dont agree why not contact Mike Chaney yourself with your concerns?

at 7/28/2007 2:58:34 AM, Walter said:
Robert, I get the feeling you''re missing our point here. Yes, all cameras do calculations to determine values for every pixel. The Foveon takes 14.1 million values, and ends up with 14.1 million values (4.7 million locations x 3 colors each) By contrast, the Canon 5D starts with 12.7 million values, and ends up with 38.1 million (12.7 million locations x 3 colors each.) This interpolation is a little different from up-ressing a picture - but the results are similar. And this interpolation is the reason you have to sharpen a Bayer image. I''d like to see you find any authoritative source that says the Sigma sharpens more. The fact is, it doesn''t - and it doesn''t have to, because the Sigma isn''t making a larger image than it has information for. Yes, it is referenced on the DPR forum - but wasn''t that the point? After all, the article is addressed to the "Foveon Fanboys" - so shouldn''t the experienced owners at the Sigma forum be given a chance to comment? The fact is, most Sigma owners understand the situation and questions better than anyone else... We''ve had Bayer cameras, and Foveon cameras, and know what hey do in the real world. Several of us have and occasionally shoot Canon pro bodies - but for some applications, they still prefer the Sigma cameras. Walter

at 7/28/2007 7:22:49 AM, George Preddy said:
Brian, I think you will just have to accept that people will use whatever appeals to their own eyes. I myself shoot with Canon, and constantly hear comments (even on this blog!) about ''plasticly'' skin tones from those cameras. I think this is rubbish personally, but I respect that my eyes are not infallible - perhaps there is something there that I cannot sense, or ''pickup''?

at 7/28/2007 7:28:53 AM, Brian Dipert said:
George, the 'plasticky' feedback you've gotten sounds akin to an argument that an audiophile would make, in justifying his or her multi-thousand-dollar speaker wire ;-) As for me, as you say with your Canon gear, I'm pretty darn happy with my Pentax and Samsung 6/10 Mpixel DSLRs. Hey, if you can't see 'plasticky' artifacts (putting aside for the moment whether or not they're even there), then why spend extra money to deal with them?

at 7/28/2007 10:01:00 AM, Doug said:
You are missing the point. The true test of a camera is its ability to produce image quality. The argument about mega pixels, is it and isn''t it and so forth, is irrelevant. Red

at 7/28/2007 12:48:18 PM, Super-Genius said:
It has 14MP. A low pass filter reduces resolution anyway. The SD14 is a stinker because the sensor is too small and noisy, the color reproduction is not professionally salable, the body is way too slow when saving images, the LCD screen is a cheap toy, the lens mount is weak, the flash metering is Neanderthal, the software is buggy, and the price is double what it should be. Other than that it's revolutionary.

at 7/28/2007 1:55:38 PM, Robert said:
Alf - I'm referring to John Clifford's test where he clearly said "A 2268x1512 center crop out of the 5D's image will be identical to the image from the SD10 in terms of image coverage." John - I write image processing software and can precisely explain how Bayer works, but this is not the place for it. Suffice it to say that your understanding is not correct. Walter - Sigma PhotoPro sharpens by default when set to 0. Common knowledge among Sigma users. Other cameras 0 sharpening means 0 sharpening. Also, the bottom layer of Foveon has a little sharpening applied separate from the SPP setting.

at 7/31/2007 3:47:49 AM, John Clifford said:
Robert, I have written image processing software, and again I ask you to tell me how you can accurately determine the color at a single 'pixel' based upon a monochromatic value... and the answer is, you can't. That is why (and because) Bayer image processing algorithms MUST look at the values of adjacent 'pixels' in order to determine what the 8- to 12-bit number indicating relative light strength actually MEANS at that location. Re the center crop test... I did this test, and found that, yes, because the 5D shared a 'pixel' diameter of almost exactly that of the SD10, then a center crop from the 5D would be almost identical in scene coverage to a full-frame image from an SD10 given the same lens. In other words, IF the SD10 was a full-frame sensor instead of a 1.7x crop factor APS-C-sized sensor, it would have the same number of 'pixels' as the 5D. Capiche? Therefore, comparing such images is valid for determining PER-PIXEL RESOLUTION, not camera-to-camera resolution. Per-pixel resolution comparisons are a valid way of evaluating the merits of Foveon versus CFA technology (as opposed to SD10 versus 5D), and the Foveon design wins as expected. The 5D will outresolve an SD10, as well as the SD14, but the latter comes a lot closer in terms of actual captured resolution (versus 'pixel count')... and the per-pixel resolution shows that, again as expected, a 2x2 multi-row panorama that provides 12 MP of Foveon sensor area greatly out-resolves a single image with the same number of MP from the 5D. In short, all of the bashing the SD14 has received on this blog is irrelevant, since Sigma acknowledges that it didn't set out to build a camera that would compete on features with the D200, 5D, etc., but one that would compete in terms of image quality... and they did. The D200 has better AF, better TTL AE, etc., but given proper exposure the SD14 has better image quality.

at 8/5/2007 7:01:22 PM, petr petr said:
Did someone made simple shot of red cardboard for example? Sigma with latest foveon should be recording 4.7mpx single colour,correct? And what about bayern system people? Try that and will see!!!!!!!!

at 10/6/2007 9:57:38 AM, Mr. Nmptlpt said:
Some like Ectachrome, some Kodachrome. I always liked Agfachrome 50. I developed my Tri-X in D76 or Acufine and printed "down with contrast," grain and all. I like the results of the SD14 and there are some wonderful programs like PhotoZoom for enlargements. Frankly, I think too many of you out there are "spoiled rotten." Compared to what we had even as little as 30 years ago, we are in the midst of a technical revolution that has rocked such giants as Kodak. The SD14 provides results that are, INM”H”O vastly superior to film, especially in the light of softwear like PhotoShop, DXO and PTGui. Just think of it, we can now buy software that makes our electronically rendered photo look like, you guessed it, Tri-X, Velvia, Sepia or Kodachrome. Some of the best photos were made with Brownies So let us (notice, I include myself) stop whining and start making Photgraphs!

at 10/6/2007 10:18:13 AM, Mr. Fanboy said:
Dear Mr. Expert, "Sigma SD14 Dissatisfaction: Finally Believe Me, Foveon Fanboys?" Why should I believe you? There are other like Mr. Mike Chaney, also an expert who might differ. Me thinks thou protest too much! Isn''t that a bit Cavalier to the point of self-appointed "guru-dom?" Yours truly, Mr. Fanboy

at 11/3/2007 9:11:50 AM, Robcad said:
To this Brian Dipert and what's the other guy's name... ah McNamara yeah. please read www.ddisoftware.com/sd14-5d/ and then realize that things are a little bit different than what you stubbornly think... Regards, robcad

at 12/16/2007 11:55:26 PM, Derek said:
It is mid Dec 2007 now, and Poor Robert is still trying to work out how to explain that: "Bayer does not guess pixels, it calculates them". Robert, if indeed you "write image processing software and can precisely explain how Bayer works" (if so, may God have mercy on us all) it should be trivial for you to answer that one. I am sure that you are a busy person, so how about a link to a place where YOUR software which can uniquely do such miracle is explained in a bit more details... any technical articles, patents, presentations, journal articles, anything? No? Great pity, for the photographic industry your method of being able to accurately determine the color at a single ''pixel'' based upon a monochromatic value (no guessing! accurately!) would be revolutionary (and for you personally it would bring a lot of rewards, including $$!).

at 1/13/2008 11:13:29 PM, Don Ferguson said:
Look at the title of this post!! And then everyone of your reponses call on popular photographers Mr. Mcanara for backup. You two should get married and live unhappily ever after. There is simply no room for the likes of you in this sofisticated world we all live in now. This type of bias simply has no place in the world where we are all trying to get to the real truth where we simply aren't going to get it from manufacturers so we rely on reviewers, but to use words like fanboys! you have shown yourself to be an idiot. Your whole take on this is simply an insult to any intellegent person! I simply hope that your publishers or superiors will usher you out the door and replace your xxx with someone of some integrity! Don Ferguson

at 4/13/2008 8:32:56 PM, Sam Murphy said:
When Sigma/Foveon first came to market, they promoted their megapixel count with reference to only the single layer count, and attempted to educate the public about the differences between foveon and bayer. Later, they began to use the full 3 layer count as they believed (and rightly so) that people were not fully understanding the complexity of the foveon architecture and belived they were getting short changed. Even the author of this article would agree that the foveon sensor resolves more than it''s single layered bayer counterpart, eg. the 4.7MP foveon would out resolve a 4.7MP bayer, however, because of naming/counting conventions, the foveon get''s short changed (from a marketing point of view). If you were to do such a comparison, the bayer sensor would be put to shame, in part because of it''s interpolation, and in part because of its sharpness robbing anti-aliasing filter. Simple mathematics explains the differences in resolution among the colours: a 10MP bayer (eg. the Nikon D80 popphoto uses) has 2.5MP blue, 2.5MP red and 5MP green. So of course, it does better with green, and worse with red and blue, which is precisely the results popphoto came up with. I agree, Sigma/Foveon should not claim a 14MP sensor, but is it fair that they be handicapped so much (popphoto, as you quoted, give the foveon 8-9 MP resolving power), simply because naming conventions do not allow for their sensor design?

at 5/17/2008 11:02:15 AM, wuzabeeb said:
Excellent discussion .. from a broadcast video background and working very much in HD at present .. it seems that the main area ignored is that the lens puts down image detail accurately on to a sensor surface that is - in the case of Foveon - doing what film does and producing single location colour accurate results .. whereas Bayer sensors are location shifting the optical image colour information .. so will produce output that is more 'testing' of the lens capability, especially as we approach higher pixel counts on finite area surfaces of sensor. We also have to take into account the 'post processing' that is essential to bring correlation of the colour information on Bayer sensors. Pixel shifting works on video cameras.. but leaves a less than satisfactory impression of 'resolution lost' along the line somewhere. I think the marketing 'hype' regarding pixel 'resolution is something that it is good to see challenged .. but commonsense tells us it's only a matter of time before the sensors exceed the resolution of 70mm film .. and then we all relax? In the meantime ... a '14' MP Foveon sensor working in 'native' mode produces a much smaller file than a 14MP Bayer sensor yet only a fraction less resolution ..according to the tests ... surely this should be the correct comparison to look at?

at 6/25/2008 6:09:18 PM, stedwards said:
maybe late to the party, I just read the popphoto report just now, I own both Canon & Sigma DSLR's for varying reasons. I found the comparison (to the Nikon D80) somewhat unfair in the choice of lens used(the Nikon used a fixed 50 1.8, the Sigma an 18-50 2.8 zoom) just guessing that it was the reason for the contrast falling low towards the corners on the sigma files. In my own experience, majority of my clients prefer product shots done with the Sigma SD9/SD14 (over the Canon 20d/1ds-M1).

at 7/3/2008 5:16:55 PM, Mr Unbiased said:
I have no bias with respect to the SD14. I'm currently lining up to buy my first DSLR having owned only a compact since my Practika 35mm kit gave up the ghost several years ago. I have read just about every article there is online re. the SD14 (including this load of baloney) and what I see it an underrated innovative technology inside a less than perfect body. I've seen loads of direct image comparisons with cameras now costing two or three times more (since the SD14 has dropped through the floor, price wise). It seems to me that the SD14 represents the closest to a good film camera that a mid-price DSLR can currently achieve. Most of the gripes about the SD14 are actually non-issues in many respects for me since I've always liked my 'serious' cameras primitive (from my Zenit EM when I was 12 - no TTL metering :) ). That said, it's a toss up between the ease of use and convenience of an Olympus E510, robustness and simplicity of a Pentax K10d (or the Samsung clone) OR the DS14. Truth is, I couldn't even consider a DS14 at its original retail price, but I suppose I should be thanking all the hacks who have dismissed it out of hand (or perhaps brand bias) for forcing the price down to a level when I can give it a try and swap it out for a K10 if I don't like it. Instinct says I'll probably like it though - though the two features I wish it had are in-body stabilization and weatherproofing. Were it not for the lack of those features the game would already be over.

at 7/8/2008 8:09:14 PM, Geoff said:
Even later to the party but I just can''t resist pointing out a few things: 1) That the SD14 is not a 14 MP camera is hardly news. Of course it''s not. 2) This statement "So, in real-world shooting -- where it''s hard to find finely detailed red-and-blue subjects or scenes that have little green or b&w detail" is entirely subjective and is best disregarded, perhaps with the rest of the article. What is "Real world shooting", and for WHO? How much is "little" detail? If you shoot sports and I shoot orchids, our "Real world shooting" will be quite different. 3) What''s the point in rubbishing a technology that is attempting to improve on the current norm? That''s my three cents.

at 9/24/2008 7:25:44 AM, Nam said:
The usual fanboys go against a neutral and unbiased reviewer. My Guess: Foveon/Sigma employees have a clausel in their contracts that forces them to read and post in forums. It reads like some posts are written by marketing managers.

Brian's Brain

Profile

RSS Feed

Recent Posts

Recent Comments

Most Commented On

Archives