PDA

View Full Version : Foveon vs. CMOS image sensor


22littlereasons
17th of June 2005 (Fri), 20:24
I read a couple articles in trade mags waxing poetically about the fantastic detail, definition and colour richness of the sigma foveon image sensor. Although it's only about just over 3MP, apparently the radically different design difference makes up the lack of MP's.

I have an acquaintence that shoots professionally with a Canon for the newspaper, and does portrait and landscape work on the side. He swears by sigma sensor performance, however, he's not too enchanted the the camera build. He also owns a fuji, along with a canon that he uses for other specific types of shooting.

Just thought I'd bring it up as the start of a discussion. I truly don't know much more about it than what I've read... yeah, yeah, I know it's a Canon forum but I don't think that means we can't expand our minds a little. *s*

ron chappel
18th of June 2005 (Sat), 05:42
Is the sigma design still about three megapixels (but marketed as equal to nine megapixels)?

Just some thoughts-

The sigma is indeed much better than it's three megapixels imply.How much better is a matter of opinion and most judge it to be worth about five-ish Mp (or was it four?...i can't rememberthe exact comments now!)
I wonder how much in camera sharpening and saturation the sigma does?

The Fuji S2 pro was in a similar situation-it had six megapixels (12 interpolated Mp) but gave detail equal to about eight-ish Mp .(I think the S1 and S3 were the same design as well)

One thing to remember is that bayer sensor cameras like the canons DON'T intepolate pixels like stigma/fovean imply- they simply interpolate SOME colour info.So it's no surprise that there is a massive difference between what sigma says and what canon owners are getting

Hopefully what i've said doesn't sound like a full on anti sigma thing.I kinda wished they had done better but it looks like they are now permanantly relegated to fringe status.
Being locked into sigma only lenses was my biggest turn off.No Jpeg ability was also silly

CyberDyneSystems
18th of June 2005 (Sat), 06:27
How old were those articles??

Either they were a few years odl,. or the authors were a little behind the times..

The Foveon tech is a great Idea.. but there has been little to no forward motion in the tech since that 3MP design was installed in the Sigma DSLR... what.. now three years ago.

The Foveon with it's 9MP of data.. created images that resolution wise were close to a 6MP CCD or CMOS,.. and of course has that brilliant color.

The funny thing is though,. that tests of color accuracy have put the Canon DSLRs at the top of the pack since the 10D.. Mind you we are talking about "accuracy" IE: reall life reproduction.. well accurate is rarely what people want. Deeper saturation and "Velvia like" color and contrast are the demand... and for that,. in a Canon DSLR post processing is required.

My point though is that if the 10D was able to provide more accurate color rendition than the SD9.. then where exactly does that leave the Foveon techs advantage?

PacAce
18th of June 2005 (Sat), 07:04
Is the sigma design still about three megapixels (but marketed as equal to nine megapixels)?
One thing to remember is that bayer sensor cameras like the canons DON'T intepolate pixels like stigma/fovean imply- they simply interpolate SOME colour info.So it's no surprise that there is a massive difference between what sigma says and what canon owners are getting

From what I understand of the bayer sensor, they, in fact, DO interpolate pixels as well as color info as they go hand in hand. The Foveon sensor, however, does NOT interpolate anything. The latest revision of the Foveon sensor has 3.4 mp of photosites. But each photosite can discretely record luminance info for each of the 3 RGB colors, thus resulting in 10.2 mp of image data.

Now, compare that with a bayer sensor which has, say 6.3 mp of photosites. Each photo site can only record luminance info for one specific color. The luminance info for the other two colors in that pixel position has to be interpolated by using info from the neighboring pixels of the same color.

So, comparing pixel counts between the Foveon sensor and the bayer sensor is like comparing apples and oranges. However, I will go so far as to say that theoretically, with everything else being equal (pixel count, manufacturer, etc.), the image quality from a Foveon sensor would be far more superior to that from any Bayer sensor. I think a Canon with a Foveon sensor would be awesome! :)

condyk
18th of June 2005 (Sat), 07:28
I really fancied a Sigma SD10 when I first started looking for a DSLR mainly because I'd got good results from my Sigma SA5 SLR and 400mm zoom in the past. The camera really seemed to produce some lovely 3D and deeply rich images ... maybe some of the very best I've ever seen. Couldn't find one at the price I wanted to pay and also started to feel that the Sigma tie in with lenses wasn't in my wider interests. I like as much choice as possible ...

I still wonder why the sensor hasn't been taken up by alternative brands and I guess the Apple/Windows debate and the Betamx/VHS debate are similar. Arguably a better technology (I say arguably!) but brand dominance is very powerful and there are fundamental vested interests. Maybe one day camera's will be so modular that we can slip in a new sensor the same as we can a new lens, flash, whatever.

twalker294
18th of June 2005 (Sat), 10:13
I think that as it stands right now, the Canons beat out the SD10 for image quality. However if Foveon comes out with a 6 or 8 mp X3 sensor, it would likely be a very different story. The shots from the SD9/10 just have a very unique quality to them that can only be attributed to the unique sensor. I poopooed them when the SD9 came out and for good reason -- the SD9 had some serious color issues. But the SD10 appears to have fixed those problems and the pictures from it are quite good. I don't like the quality of Sigma cameras but if Foveon was able to come out with a 1.6 crop X3 with 6 or 8 mp then sell it to Canon or Nikon, I'd be first in line.

Todd

Jesper
18th of June 2005 (Sat), 14:41
One thing I've read about the Foveon sensor is that it has a lot more noise than Bayer-type sensors. A Foveon sensor has the three photosites stacked on top of each other. The photosites on the bottom get less light than the photosite on top, which means the signal of the deeper photosites has to be amplified more, leading to higher noise.

At first sight the Foveon sensor sounds like a great idea, but it's not 10 MP as Sigma wants us to believe, and there are some downsides like the noise issue.

karusel
18th of June 2005 (Sat), 14:55
Ah, I think Foveon is an exotic deviation that just can't last in the fast paced consumer oriented market. It would be good if it had equal mp as the competition these days, comparably the same speed camera, comparably low noise at higher ISOs, etc, for comparably the same price. Then it would sell.

Anders Östberg
18th of June 2005 (Sat), 15:46
The sensor is only one part of the overall system. I'm afraid Foveon can be quite a bit better but still not make me switch from my current Canon gear.

ron chappel
18th of June 2005 (Sat), 16:26
I just had a look at the in depth tests at dpreview.Very interesting
Unfortunately they use a slightly different test method for the sigma (compared to their usual) but the results do look very impressive!
http://www.dpreview.com/reviews/sigmasd10/page13.asp

Note that this test is against the canon 10D ,not the latest 20D. The sigma IS quite competetive ! (i even prefer that TYPE of noise it gives compared to canon's typical blotchyness)

In the end they are both 'good enough' for most users.It comes down to other factors like price,lenses,etc

PacAce
18th of June 2005 (Sat), 17:50
One thing I've read about the Foveon sensor is that it has a lot more noise than Bayer-type sensors. A Foveon sensor has the three photosites stacked on top of each other. The photosites on the bottom get less light than the photosite on top, which means the signal of the deeper photosites has to be amplified more, leading to higher noise.

At first sight the Foveon sensor sounds like a great idea, but it's not 10 MP as Sigma wants us to believe, and there are some downsides like the noise issue.
I wonder how accurate this statement is given that film negatives and slides basically work the same way (as far as color getting through the different layers is concerned). Not that I'm doubting you, Jesper. I'm just wondering. :)

Longwatcher
18th of June 2005 (Sat), 17:52
I actually like the concept behind the foven sensor, it is without a doubt the way to improve a sensor when you get to the physics limits on the wavelength versus pixel detector size.

I don't know how well it manages noise, but based on the technical papers I read, it should provide the same or at least nearly the same SNR as bayer filter single plane sensors (ala Canon style CMOS) however you should get greater and more accurate color depth from a foven sensor. It also has the potential to get higher dynamic range by interpreting the results from each stacked sensor differently. Eventually you could as technology progresses add more detectors to the stack providing even greater color depth and possibly at the same time a true IR capability.

However, the ones I have heard about are still only a 3MP sensor from a resolution perspective. They need to get to at least 8MP and eventually preferably 22MP to reach the peak capability. So with luck maybe Sigma will hook up with Canon or heaven forbid Noink and work to get that ultimate camera out there in 8-12 years from now.

Working the programs I have worked on that doesn't seem like very long at all.

Just my opinion,

slin100
18th of June 2005 (Sat), 20:17
The idea of stacked RGB sensels looks really good on paper, but as I understand it, Sigma's implementation leaves a lot to be desired. The color separation of each channel is supposedly very poor (there is a lot of contamination between channels), and the bottom layer receives very little light. As a result color accuracy is, surprisingly, one of its weak points. It seems to have a characteristic of rendering skin as a ugly, jaundiced tone.

Another point is that the human visual system is something like 10 times more sensitive to luminance than chrominance. The Bayer sensor capitalizes on this fact by dedicating 2 green sensels for each red and blue sensel. This notion of sacrificing color resolution for luminance abounds in existing technology; NTSC signals and even JPG images are encoded with lower color resolution than luminance. The supposed advantage of stacked RGB sensels, therefore, buys very little that is perceivable. A Bayer CFA is a closer match for the human visual system.

The final point, while not really relevant to Foveon, has to do with Sigma's SD cameras. They do not use an antialias filter, which is a must have in any Analog-to-Digital system. Without it, terrible aliasing artifacts which can range from moire patterns to stairstepping can occur. Sigma SD cameras have a reputation of producing sharp images, but much of this sharpness is false. Canon DSLRs properly employ an antialias filter. While images may appear softer out of the camera, the detail is there and accurate.

Longwatcher
19th of June 2005 (Sun), 17:19
The idea of stacked RGB sensels looks really good on paper, but as I understand it, Sigma's implementation leaves a lot to be desired. The color separation of each channel is supposedly very poor (there is a lot of contamination between channels), and the bottom layer receives very little light.

I can't speak to Sigma's implementation of the Foveon sensor, but that cross contamination should in theory (within the sensor) be a good thing if it were properly used. Each channel should essential create its own layer, but the lower layers should be using the data from the layers above them in interpreting the intentisity value (luminance) of the light (as opposed to color). I am partially guessing here, but I suspect the lenses used may also have an effect on the cross contamination, by not focusing the light correctly in the first place.

As I mentioned, I like the concept and I think eventually that will be the only way to go, but not until they get it right.

Just my opinion,

Jesper
20th of June 2005 (Mon), 00:19
I wonder how accurate this statement is given that film negatives and slides basically work the same way (as far as color getting through the different layers is concerned). Not that I'm doubting you, Jesper. I'm just wondering. :)Well, film works completely different than digital sensors, so I don't think you can compare them just like that. And high ISO film indeed has a lot of grain, in fact, one of the reasons I bought a 10D is that it's better than film with regard to noise or grain at high ISO's.

PacAce
20th of June 2005 (Mon), 05:40
Well, film works completely different than digital sensors, so I don't think you can compare them just like that. And high ISO film indeed has a lot of grain, in fact, one of the reasons I bought a 10D is that it's better than film with regard to noise or grain at high ISO's.
How so are they different? A little enlightment would be greatly appreciated. Thanks. :)

ryogajyc
26th of June 2007 (Tue), 19:06
How old were those articles??
The Foveon with it's 9MP of data.. created images that resolution wise were close to a 6MP CCD or CMOS,.. and of course has that brilliant color.


I believe you are confusing MP (megapixel) with MB (megabyte). On a Foveon X3 with 3MP of sensors, each pixel records 3 color channels for RGB, each channel consuming one byte. Therefore, there's 9MB of data.

For a 9 MP Bayer CMOS sensor there is 2.25MB of red channel, 4.5MB of green channel, 2.25MB of blue channel. This is effectively 3MP worth of data, however with interpolation, they bring this up to 9MP.

Arguably a better technology (I say arguably!) but brand dominance is very powerful and there are fundamental vested interests. Maybe one day camera's will be so modular that we can slip in a new sensor the same as we can a new lens, flash, whatever.

"Arguably" implies not truly understanding the X3 technology. A Bayer sensor throws away 2/3 of the light falling on it. A Foveon X3 sensor of the same size captures all of the light (ignoring small losses intrinsic to any sensor). That's 3 times more light information captured on an equal comparison. X3 sensor technology is superior to Bayer sensor.

One thing I've read about the Foveon sensor is that it has a lot more noise than Bayer-type sensors. A Foveon sensor has the three photosites stacked on top of each other. The photosites on the bottom get less light than the photosite on top, which means the signal of the deeper photosites has to be amplified more, leading to higher noise.

At first sight the Foveon sensor sounds like a great idea, but it's not 10 MP as Sigma wants us to believe, and there are some downsides like the noise issue.

What you've read is the photo from a Foveon sensor camera appears more noisy than a Bayer sensor camera, which is an important distinction. The photo taken is a result of image capture (sensor) and digital image processing. Canon has is no doubt much more advanced in terms of digital processing algorithms and other techniques. For example, they have a noise cancellation technique where they expose the sensor for image capture, then then expose the sensor to no light to record the noise on the sensor and then cancel it out of the image captures. As Longwatcher stated, if you look at the sensors alone, the X3 sensor most likely has the same or less noise than a Bayer sensor.


The final point, while not really relevant to Foveon, has to do with Sigma's SD cameras. They do not use an antialias filter, which is a must have in any Analog-to-Digital system. Without it, terrible aliasing artifacts which can range from moire patterns to stairstepping can occur. Sigma SD cameras have a reputation of producing sharp images, but much of this sharpness is false. Canon DSLRs properly employ an antialias filter. While images may appear softer out of the camera, the detail is there and accurate.

This seems to indicate a lack of understanding how sampling works per of Nyquist-Shannon Sampling Theorem. Anti-aliasing is only necessary to smooth out the sampling artifacts if the image is not twice the resolvable DPI of whatever/whoever is viewing it. The sharpness is not false, it is the true sampled/digital representation of the image. The softer output of the Canon DSLR is a result of digital processing of the image and is false. The same could be achieved by performing the same digital processing on the output of a Sigma DSLR.

Well, film works completely different than digital sensors, so I don't think you can compare them just like that. And high ISO film indeed has a lot of grain, in fact, one of the reasons I bought a 10D is that it's better than film with regard to noise or grain at high ISO's.

A Bayer sensor works very differently than color film. However a Foveon X3 sensor works similarly to film in that the colors have to travel through several layers and the layers are sensitive to different colors. So PacAce is correct.

At this point, I should note while I do Foveon X3 is better than Bayer, a camera do not a sensor make. As previously stated, a camera is a system of which the sensor is one component. Sigma uses a superior sensor, but Canon has superior algorithms and system. I myself own a Canon DSLR. Ideally, we would see a Canon body/algorithms with Sony's IS in the body, a Foveon X3 sensor, live preview and underwater housings ala Olympus.

dekalbSTEEL
26th of June 2007 (Tue), 19:15
2 years and 6 days, wow, I think we have a new record!!

RichNY
26th of June 2007 (Tue), 20:29
'2 years and 6 days' would make a great title if the title fairy is listening :) I think it would also be a record for getting a title after their first post.

kpt4321
26th of June 2007 (Tue), 21:39
I realize the majority of this thread is old, but there is no sense in letting misinformation stand.

"Arguably" implies not truly understanding the X3 technology. A Bayer sensor throws away 2/3 of the light falling on it. A Foveon X3 sensor of the same size captures all of the light (ignoring small losses intrinsic to any sensor). That's 3 times more light information captured on an equal comparison. X3 sensor technology is superior to Bayer sensor.

A Bayer sensor does not "throw away" 2/3 of the light falling on it. All of the light which hits the photosites is used in determining the final picture, the difference between the Bayer pattern sensor and the Foveon sensor is the manner in which the light's properties are measured. How much of the incoming light information is captured depends on a whole slew of things, such as the geometry of the photosites, which are being glossed over here. Anyway, all of the usable information is captured in the form of luminescense information, while some portion of it (corresponding to the "third" that you stated) is used in determining color information.

It's also worth noting that a Bayer/Canon CMOS has roughly 3 times (in the case of the XT/20D/30D) or more the number of sensor sites than the Foveon; even if it is only getting the full color information every 3 (actually more, since it uses a GRBG filter) pixels, it has about 3 pixels to every 1 on the Foveon sensor, the the "increased light information" is a moot point. Combine the light information captured by 3 photosites by the XTi into a single pixel, and you have just ended up with essentially the same image as the Foveon; however, you also threw away a bunch of additional information with regard to the distribution of luminescense, which the higher pixel count sensor has a large advantage in, regardless of filter type, and to which the human eye is rather sensitive.

It is a cool idea, but the real-world benefits with regard to the actual competition are being greatly overstated here and elsewhere.