Do you think the image of SD14(SD15/DP1/2) is cleaner than that of SD1m(DP1m/2m) ?

Started Jan 12, 2013 | Questions
Roland Karlsson Forum Pro • Posts: 27,688
Re: A correction to my post above :-(

xpatUSA wrote:

After reading Clark's article again, "that limit" is stated quite clearly as 5um which makes the SD1, at 4.9um, almost perfect in that regard, especially around the f/4 mark. Cameras with larger pixels at that same aperture would show "jaggies" more clearly on edges.

The Clark article is simply wrong.

He first says that today´s sensors are photon limited. And then he goes on and says that increasing number of pixels increases noise and decreases dynamic range.

That is only correct if you assume that you are going to either crop the image or enlarge it so that the pixel density is the same in the print, and then also look at the image at the same distance.

Which generally is a bogus assumption.

If you print the whole image at the same size, then there will be no IQ problems with having more pixels, if you are photon limited.

Moreover - there is ABSOLUTELY not any 5 um limit. The limit is strongly dependent on the technology for the sensor. If the technology is such that the percentage of captured photons do not drop when having more pixels, then the noise and dynamic range is not affected at all. We are not there yet, but getting closer for each generation of sensors.

 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
digi2ap Contributing Member • Posts: 716
Re:YES

Interesting thread. I've not upgraded from my DP2X to a DP2M because I suspect this to be right. The DP2X images are wonderfully clean on screen and, for whatever printing I need, the DP2X can print plenty big enough. I may change my mind in time, though!

-- hide signature --
 digi2ap's gear list:digi2ap's gear list
Sigma dp1 Quattro Sigma SD1 Merrill Panasonic Lumix DMC-GH3 Sigma 10-20mm F3.5 EX DC HSM Sigma 30mm F1.4 DC HSM Art +2 more
xpatUSA
xpatUSA Forum Pro • Posts: 13,277
Re: The Clark Article is Simply Wrong

Roland Karlsson wrote:

The Clark article is simply wrong.

But Roger Clark is very well respected although sometimes a little hard to follow:

http://www.clarkvision.com/articles/does.pixel.size.matter/

He first says that today´s sensors are photon limited. And then he goes on and says that increasing number of pixels increases noise and decreases dynamic range.

Actually he says it a little differently: "there is one fundamental limit: photon counting statistics", by which he is referring to the signal-to-noise ratio. His table follows:

Photons, Noise, signal-to-noise (SNR)

9 3 3

100 10 10

900 30 30

10000 100 100

40000 200 200

If we agree that these figures are correct, it should be obvious to everyone that the sensor which counts 40,000 photons has a better SNR than that which counts 10,000. Better by a factor of two, do we at least agree on that? Indeed, if we can not agree that photon noise is the square root of the photon count, then our discussion might as well here.

(My only problem is with Clark's use of the word "limit" which is so often inappropriate in the world of photography, since it implies a solid barrier which can not be exceeded. Perhaps a photon count of one?)

That is only correct if you assume that you are going to either crop the image or enlarge it so that the pixel density is the same in the print, and then also look at the image at the same distance.

Which generally is a bogus assumption.

If you print the whole image at the same size, then there will be no IQ problems with having more pixels, if you are photon limited.

Yes, when we make images the "same size" the implication is that one of the images is re-sampled. Upward and acutance is lost. Downward and acutance is improved.

Moreover - there is ABSOLUTELY not any 5 um limit. The limit is strongly dependent on the technology for the sensor.

I absolutely agree that there is not a 5um limit. In fact, "limit" is not even an appropriate word in this context.

If the technology is such that the percentage of captured photons do not drop when having more pixels, then the noise and dynamic range is not affected at all. We are not there yet, but getting closer for each generation of sensors.

A reasonable conjecture. Meanwhile, I believe the thread is about existing sensors.

-- hide signature --

Regards,
Ted.

 xpatUSA's gear list:xpatUSA's gear list
Sigma DP2 Sigma DP1 Merrill Panasonic Lumix DMC-G1 Sigma SD14 Sigma SD15 +16 more
Roland Karlsson Forum Pro • Posts: 27,688
Re: The Clark Article is Simply Wrong

xpatUSA wrote:

Actually he says it a little differently: "there is one fundamental limit: photon counting statistics", by which he is referring to the signal-to-noise ratio. His table follows:

Yes, thats correct.

Photons, Noise, signal-to-noise (SNR)

9 3 3

100 10 10

900 30 30

10000 100 100

40000 200 200

If we agree that these figures are correct, it should be obvious to everyone that the sensor which counts 40,000 photons has a better SNR than that which counts 10,000. Better by a factor of two, do we at least agree on that? Indeed, if we can not agree that photon noise is the square root of the photon count, then our discussion might as well here.

Yes, thats also correct.

He makes two errors in the article though.

  1. He is only talking about pixel peeping. He ignore that the more pixels you have, the higher pixel density you get when printing. Then you get averaging over more pixels, the more pixels you have. And then you get back (most of) the dynamic range and S/N you lost.
  2. Even for pixel peeping, there is no limit. You can make the well depth larger. You can have some kind of clever dynamic range enhancement. Both improving S/N ratio.
-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
Raist3d Forum Pro • Posts: 37,169
Yes
2

In this case yes. This is an issue of the new Foveon. I can only hope it's something sigma can figure out to suppress in a future version of SPP.

Ad this is Iso 100 and full spectrum light!  This is why I have to shake my head at this point when people refer to Foveon as ultimate image quality for the "knowledgeable photographer. "

-- hide signature --

Raist3d/Ricardo (Photographer, software dev.)- I photograph black cats in coal mines at night...

Raist3d Forum Pro • Posts: 37,169
The lack of white marred by green/magenta foveon splotches?
1

Laurence Matson wrote:

Am I missing something? Are these 100% crops?

Apparently you are missing the green/magenta color splotches that are on what should be pure white snow (or dirty snow or whatever variant of white snow it should be).  This is not new. Even the cat you had posted in a flash shot a while back had exactly the same problem. Photos I took with a dp2m also had this problem at low Iso a bit (some shots). Lin Evans posted one of his first shots here which also showed the problem.

My dp2 virtually never had this issue at low iso  I am hoping maybe Sigma can fix this in a future version of SPP.

Each camera has its own strengths, and a general comparison like this makes little sense. The SD14 and especially the SD15 are nice for faster write times and - in the case of the SD15 - a wonderful buffer for raw. The SD1 has just spectacular detail, which translated into amazing printed images at any size up to A0. Even at smaller sizes, this is true: set the image for 720 dpi and print away.

but we are taking about image quality. Surely one would expect the new Foveon to be unequivocally better.

In terms of noise, that is more of a processing issue. Detail, however, cannot be processed in.

the issue is not the typical noise per se but the magenta/green splotches- or "Foveon noise"- the color constancy issue. Not too trivial to process out depending on the photo.

Shooting a wedding professionally and having to watch for this is a nightmare.

michaeli wrote:

I found that there are more purple and green noise in the dark area of the images of the SD1m than that of the Sd14 when both in low ISO (ISO 100). Is the noise related to the Dynamic performance or is that just little purple fringes caused by lenses?

I think SD1m should have better dynamic performance than SD14, what do you think?

Comparison:

SD14:

SD14

SD1m:

SD1m

-- hide signature --

Laurence
laurence at appledore-farm dot com
Sometimes, life is going to hit you with a brick. Don't lose faith.
Steve Jobs
http://www.pbase.com/lmatson
http://www.pbase.com/sigmadslr
http://www.howardmyerslaw.com

-- hide signature --

Raist3d/Ricardo (Photographer, software dev.)- I photograph black cats in coal mines at night...

Roland Karlsson Forum Pro • Posts: 27,688
Re: Yes
1

Raist3d wrote:

In this case yes. This is an issue of the new Foveon. I can only hope it's something sigma can figure out to suppress in a future version of SPP.

Ad this is Iso 100 and full spectrum light! This is why I have to shake my head at this point when people refer to Foveon as ultimate image quality for the "knowledgeable photographer. "

The previous sensor was optimised for daylight. It was VERY bad at tungsten.

In the Merrill sensor they made the blue layer thicker.

Its a kind of zero sum game. Improving tungsten, worsen daylight.

-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
xpatUSA
xpatUSA Forum Pro • Posts: 13,277
Re: The Clark Article is Simply Wrong

Roland Karlsson wrote:

He makes two errors in the article though.He is only talking about pixel peeping. He ignore that the more pixels you have, the higher pixel density you get when printing. Then you get averaging over more pixels, the more pixels you have. And then you get back (most of) the dynamic range and S/N you lost. Even for pixel peeping, there is no limit.
You can make the well depth larger. You can have some kind of clever dynamic range enhancement. Both improving S/N ratio.

Looks like we're approaching agreement - I do agree that down-sampling, properly done, can improve IQ.

As far as S/N ratio is concerned,  I'm not certain that well-depth can simply be increased for a Foveon sensor because it's layers have to be at specific depths:

Foveon 14X3DP1

As to "some kind of clever dynamic range enhancement", can't really respond to such speculation - sorry.

-- hide signature --

Regards,
Ted.

 xpatUSA's gear list:xpatUSA's gear list
Sigma DP2 Sigma DP1 Merrill Panasonic Lumix DMC-G1 Sigma SD14 Sigma SD15 +16 more
Roland Karlsson Forum Pro • Posts: 27,688
Re: The Clark Article is Simply Wrong

xpatUSA wrote:

Looks like we're approaching agreement - I do agree that down-sampling, properly done, can improve IQ.

Yepp

As far as S/N ratio is concerned, I'm not certain that well-depth can simply be increased for a Foveon sensor because it's layers have to be at specific depths:

As to "some kind of clever dynamic range enhancement", can't really respond to such speculation - sorry.

For Foveon, I assume both are out of the question. Foveon pixels are enough complicated already.

But ... for ordinary sensors (with Bayer filters) thats a totally different story.

Lets say that you have

R X
G B

Instead of

R G
G B

Where X is a sensor without filter. Then that pixel would be moire sensible, digging deeper in the shadows.

X could also be a dark grey filter. That could improve high light properties.

There are other solutions. One could make the detectors logarithmic by making them less sensitive when they are charged. Possible? I dont know

What I do know though is that there is lots of reseach on this topic and that consumer products generally dont use the result from such research, for some reason

-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
Raist3d Forum Pro • Posts: 37,169
Re: Yes

Roland Karlsson wrote:

Raist3d wrote:

In this case yes. This is an issue of the new Foveon. I can only hope it's something sigma can figure out to suppress in a future version of SPP.

Ad this is Iso 100 and full spectrum light! This is why I have to shake my head at this point when people refer to Foveon as ultimate image quality for the "knowledgeable photographer. "

The previous sensor was optimised for daylight. It was VERY bad at tungsten.

In the Merrill sensor they made the blue layer thicker.

Its a kind of zero sum game. Improving tungsten, worsen daylight.

Interesting take. Do you have a a particular link that explains this design choice?  Thanks Roland.

-- hide signature --

Raist3d/Ricardo (Photographer, software dev.)- I photograph black cats in coal mines at night...

MOD Kendall Helmstetter Gelner Forum Pro • Posts: 20,149
SD-1 is cleaner, but you need to expose differently

I was busy in Death Valley and missed your post originally.

In the older Foveon cameras, we tended to be able to bring up shadows a little more.

Where the Merrill sensors differ is that they have more DR than the older sensors (substantially more) but to get at it you need to over-expose a bit.  In that sort of white snow shot for example, you'd probably expose the SD-1 at +1.7.  Possibly even higher.

The the shadows would look fine.

 Kendall Helmstetter Gelner's gear list:Kendall Helmstetter Gelner's gear list
Sigma 70-200mm F2.8 EX DG OS HSM Sigma 8-16mm F4.5-5.6 DC HSM Sigma 24-70mm F2.8 EX DG HSM Sigma 50-500mm F4.5-6.3 DG OS HSM Sigma 85mm F1.4 EX DG HSM +4 more
Roland Karlsson Forum Pro • Posts: 27,688
Re: Yes

Raist3d wrote:

Interesting take. Do you have a a particular link that explains this design choice? Thanks Roland.

Nope.

I only have two "evidences" for this.

  1. The previous sensor was totally cr@p at tungsten. And looking at the raw data, the blue layer was almost empty, only noise. I have not heard about any complaints about tungsten performance for the Merrill sensor.
  2. The RAW data in the Merrill cameras has a strong blue tint, where the RAW data from the previous cameras was brownish. Note that the Merrill cameras dont have any AFE, so the bluish tint probably is a real artefact, and not due to some amplification.
-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
Raist3d Forum Pro • Posts: 37,169
Re: Yes

Roland Karlsson wrote:

Raist3d wrote:

Interesting take. Do you have a a particular link that explains this design choice? Thanks Roland.

Nope.

I only have two "evidences" for this.

  1. The previous sensor was totally cr@p at tungsten. And looking at the raw data, the blue layer was almost empty, only noise. I have not heard about any complaints about tungsten performance for the Merrill sensor.
  2. The RAW data in the Merrill cameras has a strong blue tint, where the RAW data from the previous cameras was brownish. Note that the Merrill cameras dont have any AFE, so the bluish tint probably is a real artefact, and not due to some amplification.

I would love to see the DP2M put through tungsten next to a DP2 as a curiosity for this. But to be honest if that's what happened, I think I rather have better splotch-free daylight performance.

I wonder why they removed the AFE or why they thought the AFE would be better to begin with (or maybe it is- I never quite understood that- except that it seemed to band).

Anyhow, really hoping SPP monochrome allows good B&W into the high iso.

-- hide signature --

Raist3d/Ricardo (Photographer, software dev.)- I photograph black cats in coal mines at night...
“The further a society drifts from truth the more it will hate those who speak it.” - George Orwell

Roland Karlsson Forum Pro • Posts: 27,688
Re: Yes

Raist3d wrote:

I would love to see the DP2M put through tungsten next to a DP2 as a curiosity for this. But to be honest if that's what happened, I think I rather have better splotch-free daylight performance.

Yes, me to, on both accounts. I could live with a non tungsten camera. Even Bayer cameras are really not good at tungsten, and produce best yellowish images. If you try to compensate they normally look bad.

I wonder why they removed the AFE or why they thought the AFE would be better to begin with (or maybe it is- I never quite understood that- except that it seemed to band).

The main reason for AFE is the limited dynamic range of the A/D converter. There is some information down in the noise, even below the least significant bit.

But ... the Foveon chips has quite high noise and quite low DR (contrary to what many here claim). Then AFE might only introduce problems. A shhotout between a DP2s and DP2x would be interesting.

Anyhow, really hoping SPP monochrome allows good B&W into the high iso.

It should. B&W high ISO done normally is already better than color images from Foveon cameras.

-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
Roland Karlsson Forum Pro • Posts: 27,688
Re: SD-1 is cleaner, but you need to expose differently

Kendall Helmstetter Gelner wrote:

I was busy in Death Valley and missed your post originally.

In the older Foveon cameras, we tended to be able to bring up shadows a little more.

I thought the old cameras had problems with the shadows, requiring exact or even some over exposure.

Where the Merrill sensors differ is that they have more DR than the older sensors (substantially more)

Can you point to any measurements showing this?

but to get at it you need to over-expose a bit. In that sort of white snow shot for example, you'd probably expose the SD-1 at +1.7. Possibly even higher.

The the shadows would look fine.

Snow should always be over exposed, one or even two stops.

But, you claim that normal shots also shall be over exposed some? In that case it hints at Sigma being somewhat too optimistic regarding to ISO number.

-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
(unknown member) Senior Member • Posts: 1,857
Re: SD-1 is cleaner, but you need to expose differently
1

Roland Karlsson wrote:

But, you claim that normal shots also shall be over exposed some? In that case it hints at Sigma being somewhat too optimistic regarding to ISO number.

When shooting at ISO 200 i mostly overexpose with the "Merrills"
For example this image:
Overexposed by 1 stop, highlights can easily be recovered in SPP and i don't have to push shadows.

Maceo

Johan Borg Senior Member • Posts: 2,618
Re: SD-1 is cleaner, but you need to expose differently

Roland Karlsson wrote:

Kendall Helmstetter Gelner wrote:

but to get at it you need to over-expose a bit. In that sort of white snow shot for example, you'd probably expose the SD-1 at +1.7. Possibly even higher.

The the shadows would look fine.

Snow should always be over exposed, one or even two stops.

But, you claim that normal shots also shall be over exposed some? In that case it hints at Sigma being somewhat too optimistic regarding to ISO number.

I think Kendall is maybe proving your theory with a real life example:

The older cameras had better red/brown tone detail than the Merrill, the kind of color that often exists in shadows in nature. Thus, overexposing prevents these areas to go monochrome when pushed.

I'm not sure the dynamic range is very different between the generations, my DP2 and DP2M feel rather similar in practice, but the SD15 is completely different from both of them in that it has less highlight headroom.

Roland Karlsson Forum Pro • Posts: 27,688
Re: SD-1 is cleaner, but you need to expose differently

Johan Borg wrote:

I think Kendall is maybe proving your theory with a real life example:

He can do that of course.

The older cameras had better red/brown tone detail than the Merrill, the kind of color that often exists in shadows in nature. Thus, overexposing prevents these areas to go monochrome when pushed.

That might be right. It seems like the blue layer is thicker in the Merrill sensor.

I'm not sure the dynamic range is very different between the generations, my DP2 and DP2M feel rather similar in practice, but the SD15 is completely different from both of them in that it has less highlight headroom.

As far as I know, the SD15 is optimized for color accuracy. Due to the strange RAW color space of Foveon sensors, I think that it was possible to do some tricks that extracted more dynamic range in older cameras. Those tricks offered some color accuracy.

-- hide signature --
 Roland Karlsson's gear list:Roland Karlsson's gear list
Sigma DP3 Merrill Sigma dp2 Quattro Sony RX100 III Pentax K-3 Pentax K-1 +14 more
xpatUSA
xpatUSA Forum Pro • Posts: 13,277
Re: SD-1 is cleaner, but you need to expose differently

Roland Karlsson wrote:

That might be right. It seems like the blue layer is thicker in the Merrill sensor.

Here y'are, not to scale though:



SD1

The noticeable thing about the Merrill is the reduction in transistor count, according to ChipWorks, IIRC.

-- hide signature --

Regards,
Ted
Foveon F7X3-C9110

 xpatUSA's gear list:xpatUSA's gear list
Sigma DP2 Sigma DP1 Merrill Panasonic Lumix DMC-G1 Sigma SD14 Sigma SD15 +16 more
MOD Kendall Helmstetter Gelner Forum Pro • Posts: 20,149
Re: SD-1 is cleaner, but you need to expose differently

Roland Karlsson wrote:

Kendall Helmstetter Gelner wrote:

I was busy in Death Valley and missed your post originally.

In the older Foveon cameras, we tended to be able to bring up shadows a little more.

I thought the old cameras had problems with the shadows, requiring exact or even some over exposure.

Not that I found, beyond that old adage of "expose to the right" always being a good idea for any digital camera.  With the Merrill sensor, you want to go a bit more if you want the most DR from an image.

Where the Merrill sensors differ is that they have more DR than the older sensors (substantially more)

Can you point to any measurements showing this?

Forum search.  You even posted in the thread. It was some photography website somewhere, months back.

Or just look at the DPreview info.

http://www.dpreview.com/reviews/sigmadp1/11

http://www.dpreview.com/reviews/sigmasd1/12

Even with the SD-1 testing at ISO 100 it still showed more DR than the DP-1, although it's hard to compare those two pages exactly...

Of course there is also the fact that everyone who has shot both cameras for a while says the same thing.

but to get at it you need to over-expose a bit. In that sort of white snow shot for example, you'd probably expose the SD-1 at +1.7. Possibly even higher.

The the shadows would look fine.

Snow should always be over exposed, one or even two stops.

Of course, I mean about +0.7 more than whatever exposure compensation you would apply normally for the scene.  That's why I was saying +1.7 for that scene would yield better results.

But, you claim that normal shots also shall be over exposed some? In that case it hints at Sigma being somewhat too optimistic regarding to ISO number.

It's not required if you have limited range in an image.  It's just a good general setting since most images shot during the day have full sun mixed with deep shadow, to get the most possible DR.  The metering is probably trying to be conservative (as it is in most cameras) to make sure JPG highlights do not clip.  In the end even with +0.7 you are getting better shutter speeds than with the older cameras since you use that in conjunction with ISO 200.

 Kendall Helmstetter Gelner's gear list:Kendall Helmstetter Gelner's gear list
Sigma 70-200mm F2.8 EX DG OS HSM Sigma 8-16mm F4.5-5.6 DC HSM Sigma 24-70mm F2.8 EX DG HSM Sigma 50-500mm F4.5-6.3 DG OS HSM Sigma 85mm F1.4 EX DG HSM +4 more
Keyboard shortcuts:
FForum MMy threads