Foveon

[Gary Ferguson]

What's the future hold for the Foveon sensor technology?

If it's as good as many qualified observers claim it to be then why, in a crowded market with increasingly undifferentiated products, don't we hear more about Foveon chips?

[John Sheehy]

What's the future hold for the Foveon sensor technology?

If it's as good as many qualified observers claim it to be then why, in a crowded market with increasingly undifferentiated products, don't we hear more about Foveon chips?
[a href=\"index.php?act=findpost&pid=77245\"][{POST_SNAPBACK}][/a]

Resolution?  Their newest chip, to be used in the upcoming Sigma SD14, is 4.7 MP in an APS-frame, which is generally 8 to 10MP (or even 12MP for the D2X) now for other MFRs.

Some people claim that the Sigmas have 1.4x the luminance resolution of a Bayer with the same number of photosites, but I don't buy it.  In the resolution tests I've seen, the Sigmas get inaccurate (uneven line density, spacial shifts) at about the same point as Bayer cameras do,

[later edit - that should be, at the same point that the bayers start to degrade, but they degrade in a different way.]

adjusted for pixel resolution.  We've never seen how a Foveon with a proper AA filter performs, though.  It's very easy to confuse
AA issues vs RGB colocation issues, without this third type of system.

Color separation in the existing 3.4MP chip is not especially good, as the response curves of the channels are not very steep where they overlap.  That's why you see chromatic noise in sky and water.  The RAW data from the existing Sigmas is barely saturated, compared to the color channels from a typical Bayer CFA sensor, even when white-balanced.  Color is interpolated (or maybe extrapolated).

[bongobongo]

Probably VERY bright future!

If you want to see some images taken with the newest Sigma SD14 camera then take a look at these links:

Quite amazing it is:

http://www.pbase.com/ovasilkova/sd14
Found the link above here:
http://forums.dpreview.com/forums/readflat...thread=22440159

AND

http://www.rytterfalk.com/... .../link-from-another-time-with-sd14-21-shots/
Found link above here:
http://forums.dpreview.com/forums/readflat...thread=22446182

AND

and last one for now:
http://www.rytterfalk.com/2007/03/07/link-sd14-at-a-wedding/

Considering this latest chip from Foveon is only measuring 20.7 x 13.8 mm and only has 4.6 million effective pixels...... then I belive that this Foveon image sensor will soon be seen in other brands than Sigma cameras.

Bayer sensor are not on pair with this at all.

Imagine this sensor technology used in a future pro canon or nikon camera with a larger sensor. Droool.

According to Mr. Reichmann that will happen (Foveon sensor in a pro body, and not from Sigma) sooner than laiter ... approx 6 more months to wait.

[Quentin]

I don''t think it's in the least bit "amazing".  The problem is that even if you generously assume a 4.7mp foveon chip is twice as good as a mosaic chip, its still only hitting less than 10mp, which is now entry level for dslrs.  I guess what surprises me is why there is no 10 or 12mp foveon chipped camera, which would be very interesting.  Maybe there will be, as hinted at by MR.  Until then, its a niche product for a handful of enthusiasts.

The Sigma compact could be fun though at low ISO.

Pity too about the ridiculous "14mp" advertising claims put out by Sigma.  

Quentin

[John Sheehy]

Pity too about the ridiculous "14mp" advertising claims put out by Sigma. 

[a href=\"index.php?act=findpost&pid=106585\"][{POST_SNAPBACK}][/a]

Pity, too, all the aliasing in the images.  The "detail" at the pixel level is false detail, for the most part, assuring the brain that focus has been reached, but not accurately recording the correct location of edges and points.

[JLK]

Pity, too, all the aliasing in the images.  The "detail" at the pixel level is false detail, for the most part, assuring the brain that focus has been reached, but not accurately recording the correct location of edges and points.
[a href=\"index.php?act=findpost&pid=106586\"][{POST_SNAPBACK}][/a]

Yup. And it's a pity that CFA sensors blur detail by having to interpolate color at their sensors---those 10 MP's are really "false advertising"... Ridiculous!

The sensors work differently to collect data. It's not a straight apples to apples comparison with regard to MP---and it's a shame that potential image quality is perceived to be directly correlated to MPs in the first place. I I've seen enough comparisons with fabric and fine detail work comparing the SD14 to other 8-10 MP CFA cameras to see that the luminance aliasing with the Foveon chip is not nearly as objectionable (in most cases) as the loss of color detail and resolution with the CFA chip.

The nice thing is that while CFA sensors are probably around their maximum pixel density in dSLRs, the Foveon chip can still grow to the same density. And then it will provide more information. And perhaps they'll throw an AA filter on it for John (or do like Kodak did and have an optional one). I certainly hope that Foveon is successful enough to get to this point.

[Ray]

I have to say that my understanding of the definition of a pixel, long before I bought my first DSLR, is it's a basic 'picture' element consisting of a red, green and blue (sub) element.

When I discovered that digital cameras in general were essentially monochrome in relation to pixel count and that the 'real' pixels as I understood them, were in fact interpolations, I was quite dismayed. I thought, this is some almost fraudulent sleight of hand.

The Foveon concept conforms to my general understanding of what constitutes a pixel. If my CRT monitor is set to a resolution of 1280x1024, that means there are 1280 groups of 'red, green and blue' phosphors in the horizontal dimension, and 1024 such groups in the vertical dimension.

As I understand (although I'm not entirely sure), one of the difficulties of producing a high resolution Foveon sensor (say 10 real megapixels) is one of signal-to-noise. The red and green frequencies have to pass through layer(s) of silicon, so some degree of diffusion and loss of sharpness can be expected.

On the other hand, with a conventional Bayer type sensor, there is also a built-in degree of inefficiency. For example, light falling on a 'green' pixel lets through only green light. What happens to the red and blue component of that light? Down the drain it appears.

[cricketer 1]

Probably VERY bright future!

If you want to see some images taken with the newest Sigma SD14 camera then take a look at these links:

Thanks for providing the link to the SD14 images.  They are absolutely stunning

[Ray]

Thanks for providing the link to the SD14 images.  They are absolutely stunning
[a href=\"index.php?act=findpost&pid=106689\"][{POST_SNAPBACK}][/a]

They may well be absolutely stunning. The question is, are they more stunning than the same scene taken with a 400D? We simply don't know.

These sorts of comparisons are very flawed, but it doesn't stop people from making them. Frequently on photographic forums we get questions about lens performance where someone answering the question will post an image taken with a certain lens that looks impressively sharp. We simply don't know if another lens would have produced an image even more impressivlely sharp.

[wynpotter]

There is some talk that another camera co, maybe Fuji (they developed some of the electronics used in the sd14)will take the foveon plunge with an even larger sensor in about 6-8 months. I've waited for Sigma to get up tp the 6-8mp size and was hoping for the sd14 to be the one. If Pentax or Fuji or other step up to the plate it might make an interesting fall season.
Most new technology takes some time to finds its way and in terms of developement to retail Foveon has not done that bad. Just a thought, Wyndham

[JLK]

What's the future hold for the Foveon sensor technology?

If it's as good as many qualified observers claim it to be then why, in a crowded market with increasingly undifferentiated products, don't we hear more about Foveon chips?
[{POST_SNAPBACK}][/a]

Here's an interesting tidbit from a new SD-14 user who certainly has a technical background with imaging...

Mike Chaney

[a href=\"http://www.ddisoftware.com/sd14-5d/]http://www.ddisoftware.com/sd14-5d/[/url]

[John Camp]

I've always had the feeling that people who advocate the Foveon sensor like it for it's theoretical beauty, rather than for its practicality. Since CCD and CMOS sensors already resolve at a level higher than most lenses can handle -- and since major manufacturers seem intent on pushing that resolution to still-higher levels -- what benefit would we gain by shifting to a Foveon sensor? The limitations with the more common sensors now seem to derive more from the non-electronic bits around them, than from the sensors. If the Foveon had developed first, and was now exceeding the lenses' abilities to use it, I would ask the same thing of the CCD and the CMOS. What's the point in shifting if there's no visible difference in the end product?

JC

[BernardLanguillier]

I've always had the feeling that people who advocate the Foveon sensor like it for it's theoretical beauty, rather than for its practicality. Since CCD and CMOS sensors already resolve at a level higher than most lenses can handle -- and since major manufacturers seem intent on pushing that resolution to still-higher levels -- what benefit would we gain by shifting to a Foveon sensor? The limitations with the more common sensors now seem to derive more from the non-electronic bits around them, than from the sensors. If the Foveon had developed first, and was now exceeding the lenses' abilities to use it, I would ask the same thing of the CCD and the CMOS. What's the point in shifting if there's no visible difference in the end product?

JC
[a href=\"index.php?act=findpost&pid=107092\"][{POST_SNAPBACK}][/a]

John,

It is precisely because Bayer sensor are getting close to lens resolution limits that Foveon is becoming an interesting approach.

You get more resolution with less spatial photosites.

This is the same reason why a 6000x8000 48 MP Betterlight back is significantly better than a 39 MP Phase P45, the gap being seemingly much larger than the 20% resolution gain would normally produce.

Regards,
Bernard

[Ray]

It is precisely because Bayer sensor are getting close to lens resolution limits that Foveon is becoming an interesting approach.

You get more resolution with less spatial photosites.

This is the same reason why a 6000x8000 48 MP Betterlight back is significantly better than a 39 MP Phase P45, the gap being seemingly much larger than the 20% resolution gain would normally produce.

[a href=\"index.php?act=findpost&pid=107103\"][{POST_SNAPBACK}][/a]

Exactly! The first 3.3mp Sigma Foveon-based models had approximately the same resolution as a 6mp Bayer type DSLR. That's a huge difference per pixel. Basically a 10MB file from a Foveon sensor is equivalent to an 18MB file from a Bayer type sensor.

However, can we assume that those relativities would be maintained as pixel count increases, or might the difference be even greater? Would, say a 16.5mp FF Foveon sensor have the same resolution as a 30mp FF Bayer type sensor (5x the pixel count of the first example)? One could argue that the 16.5mp Foveon sensor could actually have higher resolution on the grounds that 30mp on a 35mm sensor is already past the point of extracting more useful resolution from current lenses whereas 16.5mp has not reached that point.

One huge design limitation in the Bayer type sensor, as I see it, is that 2/3rds of the light comprising any image (on average) is blocked before it reaches the photodiodes. That seems extraordinarily wasteful to me, but unavoidable. That's what a color filter does. An efficient green filter ensures that as little as possible red and blue light passes through. Where does the red and blue light go? It's simply lost for ever, as regards real picture information. There has to be a better way   .

[Graeme Nattress]

Actually, a 3.3MP image from the Foveon camera would need to record 3.3 million * 3 samples (be they 8bit , 12bit or whatever), that's about 10million samples to equal a 6MP bayer camera, which needs to record 6 million samples, showing how woefully inefficient the Foveon system is.

[Ray]

Actually, a 3.3MP image from the Foveon camera would need to record 3.3 million * 3 samples (be they 8bit , 12bit or whatever), that's about 10million samples to equal a 6MP bayer camera, which needs to record 6 million samples, showing how woefully inefficient the Foveon system is.
[a href=\"index.php?act=findpost&pid=107113\"][{POST_SNAPBACK}][/a]

I've just tried to explain that the Bayer type sensor throws away 2/3rds of the number of photons that comprises an image. How can that be more efficient.

I'll give you a concrete example. I expose a scene at f8 and 1/125th sec with the 3.3mp Foveon sensor and a 6mp Bayer sensor, both sensors being the same size.

During the exposure, let's say one billion photons pass through the lens in both cases. The Bayer sensor excludes 2/3rds of them through its color filter array. Only 333 million photons reach the sensor.

A theoretically perfect Foveon type sensor would not discard any of those photons. The whole billion of them are distributed amongst those 10 million photo-receptors.

That's the ideal. Unfortunately, there'll be some loss in the red and green frequencies which have to pass through a layer or two of silicon which is not perfectly transparent.

Edited: I see a fundamental principle here. The Bayer type sensor cannot discard its color filter array. That's a fundamental part of its design. Each pixel is monochrome and must exclude 2/3rds of the light (on average).

The Foveon sensor also loses some light that has to pass through layers of silicon, but that's not so much a design flaw as a temporary technological obstacle to the transparency of materials. In other words, there's scope for the Foveon design to reach its full potential, and when/if it does, the Bayer type sensor will not be able to compete.

[Graeme Nattress]

Inefficient in terms of needing a 10MB file to equal the quality of what a bayer camera gets from a 6MB file.

If you look at the response curves of a bayer sensor, there's a good degree of overlap between colour filters, so the number of photons that don't reach the sensor is much less than 2/3rds.

[BernardLanguillier]

Graeme,

If we had perfect lenses, you would be right.

This whole discussion is based on the assumption that current sensors are running into the limits of lenses and that further increasing the resolution of Bayer sensors will not deliver much actual value in terms of captured detailed.

In this context, Foveon makes a lot more sense, doesn't it?

Regards,
Bernard

[Ray]

Inefficient in terms of needing a 10MB file to equal the quality of what a bayer camera gets from a 6MB file.

If you look at the response curves of a bayer sensor, there's a good degree of overlap between colour filters, so the number of photons that don't reach the sensor is much less than 2/3rds.
[a href=\"index.php?act=findpost&pid=107122\"][{POST_SNAPBACK}][/a]

It's not a 6MB file regarding the viewable image. The compressed RAW data might be only 6MB. Often it's more. When you convert and interpolate the RAW data into a picture format, whether jpeg or tiff, it's 18MB. Foveon doesn't need to do any interpolation. It's 10MB all the way.

You are right there's some degree of overlap between the color filters, but is that a flaw in the implementation of the design or an advantage in the design? Nothing's perfect, including color filters. The colors green, blue and red can be nothing more than approximations. We're dealing with a range of frequencies for each primary color here. If we divide the visible spectrum into 3 segments, the purpose of a specific primary color filter is to exclude as much as possible the other two segments.

Maybe in practice, only half the light is obstructed in the Bayer system and not 2/3rds. It's still significant and far from ideal, but it would be interesting to get some precise data on ths.

[John Camp]

I've just tried to explain that the Bayer type sensor throws away 2/3rds of the number of photons that comprises an image. How can that be more efficient.
[a href=\"index.php?act=findpost&pid=107118\"][{POST_SNAPBACK}][/a]

I don't know much about the technical aspects of sensors; I'm more curious than anything. But if you have (as in your example) a billion photons hitting the sensor, you should have enough samples at each photosite that some fairly simple math would recover the information you'd need to get accurate color...and a variety of mathematical models could look at the various returns and tune them to make them even better...But then, I don't know how many photons are hitting a light well at any one time, especially in shadow area. The number might be too small to allow a valid statistical analysis (if you only had ten photons coming in, it'd be a problem; not that 30 coming in would be much better.)

By the way, what was the hint that Michael dropped? It's been mentioned here a couple of times, but I haven't seen the actual hint. Does he think there's a Foveon coming in a major brand?

JC