[Digitalis]

then Sigma may want to sell off Foveon Inc. I imagine there are better partners than Sigma to exploit the technology.

I heard that some time ago the tried to sell Foveon sensor technology to other sensor manufacturers but none of them showed any interest because of the way they marketed it as being "superior" to what other sensor manufactures were already making - The Foveon sensor technology isn't superior to anything else in the market, it's just different.

[JohnBee]

I heard that some time ago the tried to sell Foveon sensor technology to other sensor manufacturers but none of them showed any interest because of the way they marketed it as being "superior" to what other sensor manufactures were already making - The Foveon sensor technology isn't superior to anything else in the market, it's just different.

If I understood correctly, I think the foveon technology is superior to bayer sensor sensors in principle. However, the level of R&D applied to the technology is what is holding it back. Which I suppose could likely be true given that we've seen similar patterns with numerous other technologies.

Anyways, I found these interesting images and blurbs that I found to be very informative toward the technology.


This image above shows the essential characteristics of the Foveon chip and how it captures color.

Foveon uses three layers of photosensors. Which capture 100% of the RGB whereas a mosaic sensor captures only 25% of Blue and Red, with 50% of Green(due to green sensitivity).

However... since CCD and CMOS are photons based processes, they are essentially grayscale devices with a colored mosaic's placed in front of them. And the color is digitally interpolating into each color channels by calculating and using neighboring data. Which essentially borrows color from the pixel next door. Which in turn averages results. Which is why a 10mp bayer sensor doesn't actually equate a 10mp image.

The Foveon sensor on the other hand, isn't the result of interpolation and each photosite is the result of discrete RGB data.

Which is the reason the mosaic sensors cannot achieve their quoted resolution and the Foveon's X3 chip is considered more advanced in this respect.

In reality, I'd say the SD1 is showing some very good potential as I was looking over the following file: http://www.sigma-sd.com/SD1/sample-photo/img/Kendall-Gelner-SD1-SDIM0180.jpg

To which I'd ask... aside from the obvious staircase jaggies(lines), what other camera could deliver this type of detail?
Without looking, I'm thinking something in the lines of a D3x.
Anyways, we really need some RAW files and better examples to scrutinize this thing. But at 7000USD, I'm really not feeling it!

[Pentaxor]

If I understood correctly, I think the foveon technology is superior to bayer sensor sensors in principle. However, the level of R&D applied to the technology is what is holding it back. Which I suppose could likely be true given that we've seen similar patterns with numerous other technologies.

Anyways, I found these interesting images and blurbs that I found to be very informative toward the technology.


This image above shows the essential characteristics of the Foveon chip and how it captures color.

Foveon uses three layers of photosensors. Which capture 100% of the RGB whereas a mosaic sensor captures only 25% of Blue and Red, with 50% of Green(due to green sensitivity).

However... since CCD and CMOS are photons based processes, they are essentially grayscale devices with a colored mosaic's placed in front of them. And the color is digitally interpolating into each color channels by calculating and using neighboring data. Which essentially borrows color from the pixel next door. Which in turn averages results. Which is why a 10mp bayer sensor doesn't actually equate a 10mp image.

The Foveon sensor on the other hand, isn't the result of interpolation and each photosite is the result of discrete RGB data.

Which is the reason the mosaic sensors cannot achieve their quoted resolution and the Foveon's X3 chip is considered more advanced in this respect.

In reality, I'd say the SD1 is showing some very good potential as I was looking over the following file: http://www.sigma-sd.com/SD1/sample-photo/img/Kendall-Gelner-SD1-SDIM0180.jpg

To which I'd ask... aside from the obvious staircase jaggies(lines), what other camera could deliver this type of detail?
Without looking, I'm thinking something in the lines of a D3x.
Anyways, we really need some RAW files and better examples to scrutinize this thing. But at 7000USD, I'm really not feeling it!

this is also my understanding of the principle. and I do somehow agree to the extent that the technology might not be very well be fully utilized for whatever reason is I don't know. maybe Sigma is holding it back or hoarding it for future ventures on future cameras, thus deliberately limiting it's capabilities. that's why probably we have less than a 100% each of the RGB independently captured in each pixel. maybe Sigma just didn't research more about it or still doesn't know how to implement it fully. maybe the lack the resources to do so.

I do agree with the premise that RAW samples are needed and better examples as well. from a business and consumer standpoint, the price is more of a death warrant to the company.

[Aku Ankka]

I thought you said you have better resources? I'm asking for more basis aside from what you possibly borrowed from theories of others that coincidentally don't have the SD1 for testing such theory is indeed holds true.

Now it sounds like you think that one should reinvent the wheel and create one's own unique theories instead of learning from existing knowledge? What a weird approach to science. Nothing I've said about the SD1 sensor is of my own invention, but of common scientific knowledge.

I do know the basic scientific facts which the Foveon sensor has to obey. There is absolutely no need for one to have SD1 to know some basic properties of the sensor. No matter how much you dislike it, the laws of physics are obeyed by this sensor. It is a three layer sensor where color separation is done by the photon absorbion property of silicon. How deeply each photon travels before it is absorbed depends on the wavelength and is a propabilistic function. This is a scientific fact and no SD1 is needed for that. The color separation is weak, there will be color inaccuracy issues the cameras with organic filters don't have (to anywhere near the same degree) and so on.

And as I said, what is not known is some technical details, like if CDS is there for this sensor or not.

I see little point in continuing this conversation with you as it seems very fruitless. To me it seems that you disregard scientific realities because I don't have SD1 in my hands or whatever.

[Aku Ankka]

If I understood correctly, I think the foveon technology is superior to bayer sensor sensors in principle. However, the level of R&D applied to the technology is what is holding it back. Which I suppose could likely be true given that we've seen similar patterns with numerous other technologies.

Unfortunately this is not the case. The problem with Foveon is in the whole idea of doing color separation by taking advantage of photons penetrating material to different depth depending on their wavelength. There at least four problems with this approach.

1. The color separation will be weak leading to high noise in color images
2. The colors produces differ radically from how we humans see color
3. The sensor will be complex - the extra circuitry will lower the QE (it's 9 transistors per pixel for the current Foveon chips)
4. More data will have to be processed for the image

Foveon uses three layers of photosensors. Which capture 100% of the RGB whereas a mosaic sensor captures only 25% of Blue and Red, with 50% of Green(due to green sensitivity).

The marketing material presents somewhat false picture. While it has three photodiodes for each pixel, buried in different depths, they do not record red, blue an green. Wikipedia has a simple image of the problem: File:Absorption-X3.png - Wikipedia, the free encyclopedia

Also, the Bayer filtering sensors do not capture quite as simply as 50/25/25, but there is significant (intentional) overlap. The DxO sensor analysis pages have detailed information on this.

The Foveon sensor on the other hand, isn't the result of interpolation and each photosite is the result of discrete RGB data.

Each spatial photosite does indeed provide three measurements of the signal, but they have little to do with red, green or blue. While there is need for demosaicing with the Bayer filter images, the Foveon also requires quite heavy processing to get even remotely accurate colors out of the data - this leads to significant increase in noise, which also reduces effective resolution when the signal is low.

Anyhow, over the last few days Eric Fossum and Joseph Wisniewski have written some on the Foveon technology in the DPR - it's a good read for the curious mind.

[Digitalis]

To which I'd ask... aside from the obvious staircase jaggies(lines), what other camera could deliver this type of detail? Without looking, I'm thinking something in the lines of a D3x.

The nikon D3X has an AA filer- I think the Leica M9 would be a better choice when comparing sensor's RAW output- and since the SD1 is selling in the same price category as the Leica M9 I think it would be fair. Though optically comparing sigma Vs leica isn't even remotely fair the 70mm f/2.8 may be sharp but the Leica Summilux-M 75mm f/1.4 raises the bar on optical quality considerably.

[Pentaxor]

Now it sounds like you think that one should reinvent the wheel and create one's own unique theories instead of learning from existing knowledge?

not to reinvent the wheel, but how to effectively and efficiently optimize a "DIFFERENT" power source.

---

Last edited by Pentaxor; 05-29-2011 at 08:59 AM.

[falconeye]

Just my 2 cents too ... I don't consider the Foveon technology to be superior.

It may have a better spatial and worse color separation compared to Bayer. But spatial Bayer and color Foveon separation can both be improved by software and in all fair comparisons, should (Bayer images are not always processed this way before comparison though).

My main concern however is that both approaches don't solve the fundamental problem that a color CMOS sensor has much worse QE than a monochrome one which today can be close to 80%.

The only promising approach to solve this I know about is to replace a sensel's microlens by a trichroic microlens assembly and the underlying detector by three of them. But this approach is closer to the Bayer approach as it focusses on detecting as many photons as possible w/o caring about their property.

[Aku Ankka]

Just my 2 cents too ... I don't consider the Foveon technology to be superior.

It may have a better spatial and worse color separation compared to Bayer. But spatial Bayer and color Foveon separation can both be improved by software and in all fair comparisons, should (Bayer images are not always processed this way before comparison though).

My main concern however is that both approaches don't solve the fundamental problem that a color CMOS sensor has much worse QE than a monochrome one which today can be close to 80%.

The only promising approach to solve this I know about is to replace a sensel's microlens by a trichroic microlens assembly and the underlying detector by three of them. But this approach is closer to the Bayer approach as it focusses on detecting as many photons as possible w/o caring about their property.

I've never heard of trichroic microlens developement - I'd be delighted if you could provide a link or two on the subject as I'm very interested in sensor related developements.

Anyhow, on the other hand the problem with such approach is the inefficiency regarding amount of data and complexity of the sensor. There would be three times more data, somewhat more crosstalk due to smaller pixels and much more circuitry blocking light. If we were to use back side illumination like today for large sensors, the QE would go up by less than one could hope due to extra metal on the chip. Also one would have to create a rather complex microlens array as the light would still have to be guided through the aperture of the (sub)pixel after it's gone through the trichroic microlens, otherwise the QE will collapse totallly. I see economical problems.

Indeed this kind of technology would seem to fit BSI better than FSI as the extra circuitry would not cause problems and the QE loss for passing the (sub)pixel aperture would be much smaller.

So in my opinion, this sounds interesting, but I don't think we'll be seeing it in SLRs - possibly some variant of it in small chip cameras though where sensitivity has always been a problem, but not anytime soon if ever. Instead, I'd bet on organic photodetectors being the next big step, though that'll take at least a decade.

[falconeye]

I've never heard of trichroic microlens developement - I'd be delighted if you could provide a link or two on the subject as I'm very interested in sensor related developements.
...
So in my opinion, this sounds interesting, but I don't think we'll be seeing it in SLRs - possibly some variant of it in small chip cameras though where sensitivity has always been a problem, but not anytime soon if ever. Instead, I'd bet on organic photodetectors being the next big step, though that'll take at least a decade.

I can't find the link now. It is in a patent filed by Sony, Nikon or Canon, cannot remember exactly.

The main idea of the invention is to make every pixel a 3CCD camera, so to speak. The loss of light would have to be minimized by the microlens design. I.e., the microlens would project ALL light onto 1/4 of surface where the color would decide which fourth.

I can't see how organic photodetectors will help. They are meant to be cheap, not good. The QE of silicon photodetectors is 80%, so this isn't the problem anymore. Ideally, a pixel would be so tiny it can measure the single photon event (at least for low levels of light). The single photon event delivers accurate color information via its energy deposed (excess energy in the created electron which could be measured by some avalanche effect). But I see none of this in the literature yet.

Organic photodetectors will enable low cost large sensors. But a large sensor doesn't help in itself. It only helps by enabling a large usable glass surface of a lens' front element and making it large will never be cheap.

[Aku Ankka]

I can't find the link now. It is in a patent filed by Sony, Nikon or Canon, cannot remember exactly.

The main idea of the invention is to make every pixel a 3CCD camera, so to speak. The loss of light would have to be minimized by the microlens design. I.e., the microlens would project ALL light onto 1/4 of surface where the color would decide which fourth.

Yes, this is about how I understood it, so it sounds like better suited for BSI sensors due to the extra circuitry need, though I'm not sure, since it seems like with such system crosstalk would be a major issue to deal with, though I guess having a one more carefully adjusted microlens (micromicrolens? ) on top of the subpixel would help somewhat. Though, no knowing the details better, this is just speculation.

I've been toying with the idea of prismatic color separation for quite a while:

Big microlens -> prism -> subpixel microlens -> photodiode. Interestingly the subpixel microlens could maybe be eliminated if FSI is used and replaced with well designed light guides.

I can't see how organic photodetectors will help. They are meant to be cheap, not good. The QE of silicon photodetectors is 80%, so this isn't the problem anymore. Ideally, a pixel would be so tiny it can measure the single photon event (at least for low levels of light). The single photon event delivers accurate color information via its energy deposed (excess energy in the created electron which could be measured by some avalanche effect). But I see none of this in the literature yet.

Organic photodetectors will enable low cost large sensors. But a large sensor doesn't help in itself. It only helps by enabling a large usable glass surface of a lens' front element and making it large will never be cheap.

Organic photodetectors can be - in the future, not now - good for stacked system a la Foveon, but without the color accuracy and separation issues. At the moment they're not too good as you said, but in 10+ years the situation may be drasticly different.

Also, I wonder about the color separation process mentioned on top and observer metameric failure - to me it seems like that Foveon issue would be repeated. Then again, I am a layperson, so what do I know

(Also I am worried (yes, I do worry about way too many things) about narrow band light - without overlap in the detectors colors, this coukld be a problem?)

Anyhow, regarding that ideal situation you described, we'd have another problem in our hands, the unceretainty principle (I guess sacrificing a tiny bit of spatial and/or color information would be acceptalbe if we ever get that far though .)

[mithrandir]

This attempt by AMP to produce HDR Video would be a wonderful approach for Pentax DSLRs of the future.

Vimeo

[Digitalis]

This attempt by AMP to produce HDR Video would be a wonderful approach for Pentax DSLRs of the future.

just when I though our eyes has suffered enough from badly done still HDR photography, now we have to put up with it from the motion picture industry as well?

[Class A]

just when I though our eyes has suffered enough from badly done still HDR photography, now we have to put up with it from the motion picture industry as well?

+1

(blah, blah, blah <- required by the post submission checker)               
                
                
                
                
                
                
                
                
                
                

[mithrandir]

The problem is not the HDR imaging, the problem is tone mapping operators. Tone mapping algorithms could be developed to be both realistic and a return to the earlier Pentax look.