Future of Foveon: Serious Question

Status
Not open for further replies.
I think the strength of Foveon in the future will be in smaller cameras (like the DP series already). A DP camera with exchangeable lenses would be fine.
Yes, I would LOVE to see one! I could just put my SA mount lenses on it and focus in live view mode and every image would be as clear and sharp as possible, just like my DP2 except with lenses such as the 70-200 F2.8 OS!

Easily worth $1000 to me for the reduced hassle!
--
Tom Schum
 
A lensless DP2 that uses SA-mount lenses with (or without) an adapter would boost sales of SA mount lenses in the enthusiast crowd... I would buy one, at least. That's one sale for sure!

This would prolong the life of the most successful Foveon, the second generation one, and prime the photography enthusiast crowd for the later inclusion of the third generation sensor in the same body.
--
Tom Schum
 
greetings

The SD1 is the new Sigma Flagship camera, what's going to happen when the new Sony, Nikon, Canon, Olympus m4/3 pro come to market very soon? I'm sure they are watching what's going on with Sigma. They won't make the same mistake in pricing and get the same backlash as Sigma has. I wonder if Sigma can bring out a lower priced SD16 in time like 6 months, or a better DP series with interchangeable lenses in 6 months at lower prices to compete. And what about the lawsuit Nikon has against them? Would Nikon want Foveon in the settlement if they were to win? Nikon needs it own sensor.

Enjoy

Roger J.
 
Normally, the spin off of the "flagship" is the same camera fewer features, for example the Nikon D3 and D700. Unfortunately the SD1 is not feature laden. If it is true that the sensor is the driving cost of the SD1 to drop the price to 2K - they would have to drop back to the old sensor which would be a glorified SD15. I'm not sure Sigma has that option.
greetings

The SD1 is the new Sigma Flagship camera, what's going to happen when the new Sony, Nikon, Canon, Olympus m4/3 pro come to market very soon? I'm sure they are watching what's going on with Sigma. They won't make the same mistake in pricing and get the same backlash as Sigma has. I wonder if Sigma can bring out a lower priced SD16 in time like 6 months, or a better DP series with interchangeable lenses in 6 months at lower prices to compete. And what about the lawsuit Nikon has against them? Would Nikon want Foveon in the settlement if they were to win? Nikon needs it own sensor.

Enjoy

Roger J.
--
Truman
http://www.pbase.com/tprevatt
 
Normally, the spin off of the "flagship" is the same camera fewer features, for example the Nikon D3 and D700. Unfortunately the SD1 is not feature laden. If it is true that the sensor is the driving cost of the SD1 to drop the price to 2K - they would have to drop back to the old sensor which would be a glorified SD15. I'm not sure Sigma has that option.
agree, and the D700 was a huge sucess for Nikon (I love mine :) )
 
I think we are getting quite close to resolution limits. A 40 MPixel CFA is probably putting you at the point of significantly demising marginal return. However, in such a sensor the lens will act as an AA filter and the issues with AA filters will disappear.

The next challenge is not resolution - it is dynamic range. Back in the 70's when I was working to try to use optics for wideband signal processing, we ran into dynamic range issues with the CCD detector arrays. The dynamic range of CCD detector arrays have gotten better but not by a whole lot and CCD dynamic range is better than CMOS which is why the medium format backs, Leica, etc. use CCD.

High dynamic range and low noise has always been a tough nut to crack in digital optical arrays and still his. There are ways to get better dynamic range but they don't really apply to commercial photographic systems. Fujifilms S5 had a unique sensor design for increased dynamic range but it required two detectors per pixel hence a sacrifice of resolution for 2 more stops of DR.

If history is any indicator the DR nut will be a lot harder to crack than the resolution nut.

--
Truman
http://www.pbase.com/tprevatt
 
I think we are getting quite close to resolution limits. A 40 MPixel CFA is probably putting you at the point of significantly demising marginal return. However, in such a sensor the lens will act as an AA filter and the issues with AA filters will disappear.
You are forgetting the benefits of over sampling. There is a reason why A/D converters for sound over samples maybe 16 times. Its because you then can build advanced digital filters that makes optimal recordings. The same goes for images.
The next challenge is not resolution - it is dynamic range.
Yes - I would like more dynamic range.
Back in the 70's when I was working to try to use optics for wideband signal processing, we ran into dynamic range issues with the CCD detector arrays. The dynamic range of CCD detector arrays have gotten better but not by a whole lot and CCD dynamic range is better than CMOS which is why the medium format backs, Leica, etc. use CCD.
High dynamic range and low noise has always been a tough nut to crack in digital optical arrays and still his. There are ways to get better dynamic range but they don't really apply to commercial photographic systems. Fujifilms S5 had a unique sensor design for increased dynamic range but it required two detectors per pixel hence a sacrifice of resolution for 2 more stops of DR.

If history is any indicator the DR nut will be a lot harder to crack than the resolution nut.
Yes it will. There are solutions though. And the Fuiji is not one of the best ones. Its actually rather poor as you get lots of noise in the critical medium gray area.

I think that some day we will se more advanced solutions for DR. Solutions where you actually can get unlimited DR just by changing some parameters. Today you can design e.g. 14 stops or something like that if you are very careful- like the sensor in the K-5. But - its absolutely not impossible to get 20, 30, 40 stops ... if you prioritize DR over all other properties.

You could put a 32 bit counter in each pixel and count the incoming photons. That would give you 30 stops. Its possible to implement today. Unfortunately it will be too many transistors if you want 10 MPixels. At least today. But tomorrow?

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
 
I think we are getting quite close to resolution limits. A 40 MPixel CFA is probably putting you at the point of significantly demising marginal return. However, in such a sensor the lens will act as an AA filter and the issues with AA filters will disappear.
You are forgetting the benefits of over sampling. There is a reason why A/D converters for sound over samples maybe 16 times. Its because you then can build advanced digital filters that makes optimal recordings. The same goes for images.
The only benefit to oversampling is to allow a more gradual rolloff of the anti-aliasing filter. Analog filters (optical and electronic) have a nasty phase response on filters with a fast rolloff. You allow a slow rolloff and these nasty phase properties are mitigated. However, a 25% increase above Nyquist is quite sufficient. Much above that - you are not capturing any additional information.

Tonal smoothness is more an issue of the dynamic range of the sensor and small quantization error (more bits) than gross over sampling.

--
Truman
http://www.pbase.com/tprevatt
 
Interesting to me is the decision by Sigma to use Foveon to continue the SA lens/camera line. Was it because this was the only sensor avaible to Sigma? Was it cheaper? Where the others blocked by contracts? Is there the possibility so use more than one sensor vendor?
For me it is now about the lenses. My SA-mount lenses fit only Sigma cameras.

If Sigma changes over to a standard Bayer sensor, I would still buy it because my lenses fit. At this point, I don't see Sigma doing this, and I hope to see better and/or more affordable Foveon sensors...
--
Tom Schum
 
The only benefit to oversampling is to allow a more gradual rolloff of the anti-aliasing filter. Analog filters (optical and electronic) have a nasty phase response on filters with a fast rolloff. You allow a slow rolloff and these nasty phase properties are mitigated. However, a 25% increase above Nyquist is quite sufficient. Much above that - you are not capturing any additional information.
You are right that the main reason for oversampling is that you cannot make good analog filters.

But having a good filter is VERY important. And to get a good filter you cannot just oversample with 25%. You need to oversample several hundreds of percent. Thats why you oversample e.g. 16 times for audio.

Then - after the over sampling and running the digital filter - then you can save that data at 25% over the cut off frequency of the system. But to be able to get a nice cut off you have to over sample a lot.

Its all a matter of cost though. So - if you think sampling at 25% over cut off is good enough - then thats it. But dont fool yourself that over sampling with 1500% would not mean a much better image.

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
 
You are right that the main reason for oversampling is that you cannot make good analog filters.
But having a good filter is VERY important. And to get a good filter you cannot just oversample with 25%. You need to oversample several hundreds of percent. Thats why you oversample e.g. 16 times for audio.
It depends on the frequency. At a couple GHz you can build filters with good phase response at 25%. The lower you get in frequency the more difficult that task becomes. It all gets down to resonance frequencies of components of the filters. Also good sharp cutoff filters at audio frequencies are huge (size). The solution there is to simply use filters with a very gradual roll off as anti-aliasing filters. The sample rates are low enough in won't matter. That is not necessary at higher frequencies.

In reality we are getting to the point where there will be no need for anti-aliasing filters in digital photography - especially at the high end. The M9 has none. The medium format backs have not. The Foveon never had one - although the SD9/10/14/15 could have used one. In most cases the resolution is sufficiently high that the lens functions as an AA filter and it has much better behavior than lithium niobate which is normally used as AA filters.

There is no information above the sensors resolution. All the spatial frequency components above Nyquist are folded back over as distortion. Over sampling buys you nothings in the case the lens is you AA filter.
Then - after the over sampling and running the digital filter - then you can save that data at 25% over the cut off frequency of the system. But to be able to get a nice cut off you have to over sample a lot.
--
Truman
http://www.pbase.com/tprevatt
 
There is no information above the sensors resolution. All the spatial frequency components above Nyquist are folded back over as distortion. Over sampling buys you nothings in the case the lens is you AA filter.
You are right, but in order to make the lens a good enough AA filter you have to oversample. And this oversampling is not 25% over the resolution of the lens. If its 1500% I dont know. Maybe its only 300%. But its absolutely not 25% There are things happening with the image far above resolution limit.

You can store the image at 25% over chosen cut off frequency. But ... you will need to do some digital filtering before doing that storing. And the frequency of that filter needs to be higher.

Thats how audio works - and thats how images work - thats how every sampling system works where you cannot make perfect analog AA filters.

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
 
Kendall Helmstetter Gelner wrote:

I'm using a Sony NEX-5, which at about 1/20 of camera investment, I consider a point and shoot.
I think you might have a different definition than me for P&S, to me that's a little beyond that category... but it's good to know it gets good results.
The thing is, you are treating one vector only - high ISO support - as THE single reason why a camera will or will not survive in the market.
No really I used ISO as one example. Do you not consider the gains in resolution to be "major"
Not really, they have had some pretty good jumps but not what I would consider major, especially since in a lot of cameras gains of resolution has been offset by noise or other processing issues. Only recently have the sensors seemed to catch up quality wise with the resolution they were capturing. Along the way has been a pretty incremental climb.
And the adition of video, whether you want it or not, has created an entire new revenue stream for many wedding photographers. I think they would consider that a pretty major advancement
Sure, but to me that's kind of a specialized niche use. Serious videographers I do not think are threatened by DSLR video. If given the choice I'd still take a RED unit over any modern DSLR with video.

Also I really feel like shooting video and shooting stills are very distinct skills.
I personally would rather have a camera that produces better low ISO results, that is my own primary vector.
You can have both you know.
Not that I have seen.
I think you are new here so what I think you do not understand is that the family that owns Sigma desires to be a camera company at the level of Canon or Nikon. Indeed in part I think the current pricing reflects that as much as reach might exceed grasp... but it's further proof of the vision they have at work, that means they will be producing cameras for a long time to come, with the inevitable improvement that results.
Well I certainly disagree----I don't see the same vision you do for the company
It's not my vision, it's thiers.

Come to the next Photokina or PMA and ask yourself.

BTW, thanks for the comment about the galleries but I don't update them as often as I should these days... I feel like my newer stuff is much more interesting.

--
---> Kendall
http://InsideAperture.com
http://www.pbase.com/kgelner
http://www.pbase.com/sigmadslr/user_home
 
There is no information above the sensors resolution. All the spatial frequency components above Nyquist are folded back over as distortion. Over sampling buys you nothings in the case the lens is you AA filter.
You are right, but in order to make the lens a good enough AA filter you have to oversample. And this oversampling is not 25% over the resolution of the lens. If its 1500% I dont know. Maybe its only 300%. But its absolutely not 25% There are things happening with the image far above resolution limit.
What? If the lens only supports a spatial frequency resolution that is less than supported by the sensor - you are done. The lens is an AA filter because it attenuates frequencies below Nyquist.

In reality you can compensate for phase distortion near the band edges. You can completely reconstruct the value at any point between samples of band limited signals/images (Fourier Shift Theorem) http://www.cs.unm.edu/~williams/cs530/theorems6.pdf

The sampling theorem is nothing more than the expression of the Fourier transform of a bandlimited image ( and all digital images with or without AA filters are bandlimited to Nyquist) as a Fourier series with the samples as coefficients.

You can spatially move around in the image to any location between pixels using the Fourier shift theorem. If you have the phase response of the AA filter you can remove it. All the information available is there without grossly oversampling.

What is not there is the amplitude information lost by the dynamic range of the sensor and bit depth of the ADC. If you only have two stops - you only have two stops. Most digital cameras - including the ASP-C and 135 format offerings from all the makers including Sigma come in the 8 to 9 stop region.

Sure Nikon plays games with active fill in the D3 and a 14 bit ADC but in reality it might just buy you a 10th stop but I doubt it. The medium format sensors - because they are CCD instead of CMOS run a stop or two better.

Dynamic range is the next frontier - and it will be a tough one especially for CMOS. For CMOS to crack the 55 dB range will be a major breakthrough and for CCD to crack the 65 dB threshold will be a major breakthrough. That is where those respective thresholds have been since about 1990.

The reason high speed super computers and fully digital implementations have replaced the optical processing system of old for such things as imaging radar is not cost but the limited dynamic range of digital sensors that were used in the optical systems.
You can store the image at 25% over chosen cut off frequency. But ... you will need to do some digital filtering before doing that storing. And the frequency of that filter needs to be higher.

Thats how audio works - and thats how images work - thats how every sampling system works where you cannot make perfect analog AA filters.

--
Roland

support http://www.openraw.org/
(Sleeping - so the need to support it is even higher)

X3F tools : http://www.proxel.se/x3f.html
--
Truman
http://www.pbase.com/tprevatt
 
If Sigma were to make a 10-15MP Foveon DP3 with LV, have a hi-Rez rear LCD, offer decent shot-to-shot speed and file processing, and maybe offer interchangeable lens option, and price it at around the same as Sony's NEX, or the Olympus EP, or even a bit more $$ like the Fujifilms flawed but widely successful and highly sought after X-100, I, like many others here, would most certainly snap one up in a heartbeat. It might even become one of the best selling compact cameras in the digital era.
I like that idea, I just bought an NEX-5 for a carry around camera and it does certain things very well...plus really good video. I'll probably get one of these small cameras every few years (kids get the old technology to learn on) An NEX type camera would be cheap enough for people to try the sensor...but they HAVE to start including comparable features in any camera they sell
 
And here is the serious answer to the serious question.
For a serious camera - one I mean worth 6 K plus for the body - you better have a 135 format sensor. I think come January and the next release of Canon/Nikon/Sony if you want that much money for a camera you better be FF. The Nikon D700 is FF at about 2500 and a kick butt camera. Throw a D3x sensor in it - call it the D800 for 2500 - it is going to put a lot of pressure on cameras that want 6K plus. I expect Canon and Sony will also have a competitor.
3. A FF or even MF camera for $8000. NOTE - this means designing new lenses. Or maybe cooperating with some other MF camera makers.
Sigma has to show they can migrate the Foveon design to a 24x36 sensor before they can even consider medium format. Given they seem to have issues even going to ASP-C that I think is not likely soon.
But ... you also have to take into consideration that DSLR cameras may also be in danger. Maybe mirror less cameras is going to take over.
The reflex camera solved a problem when they were designed. Today there are other ways to solve that problem which doesn't require a reflex camera. Eventually I expect the reflex camera will give way to another design - one that is physically more compact.

I'm not sure they helps or hurts Sigma. Given Sigma has yet to produce a camera with Live View - I expect they have a lot of development to do.

--
Truman
http://www.pbase.com/tprevatt
 
Yes,

The Nyquist frequency reemains the same regardless of sampling rate. To avaoid aliasing, you just always have to sample at twice the highest spatial frequency at low pass appropriately.
 
Status
Not open for further replies.

Keyboard shortcuts

Back
Top