Luminous Landscape Forum

Equipment & Techniques => Medium Format / Film / Digital Backs – and Large Sensor Photography => Topic started by: larkis on September 25, 2011, 03:46:11 AM

Title: CCD and CMOS
Post by: larkis on September 25, 2011, 03:46:11 AM
Could someone give me a technical explanation why CMOS can shoot at such high ISO's (nikon D700, etc) compared to CCD's despite CCD's being more sensitive to light and being a more expansive technology ? A friend of mine is a scientist and tells me that in astronomy and industrial applications the reverse of what is going on in the camera business is true, CCD's are more light sensitive and have less noise compared to CMOS. I noticed with most cameras that use CCD's (like the phase backs) anything over 800iso is not really considered to be great. The sensors that go to space are also CCD's.
Title: Re: CCD and CMOS
Post by: ondebanks on September 25, 2011, 04:22:47 AM
Your friend is wrong - or at least, quoting out of date information. CMOS has overtaken CCD in noise performance and is now beginning to displace CCDs in research instrumentation. I'm an astronomer too; my collaborators have just bought some new cameras for use in a high speed photometer on one of the world's largest telescopes - and guess what, they are CMOS (Andor S-CMOS), not CCD. Future space sensors will be CMOS too; just remember that space missions often take 10 years to develop, and usually deploy very old, reliable, tested-to-death, radiation-hardened technology, so it takes a while for new tech to work through to launch.

Excellent High ISO performance requires 3 things: (1) Efficient light capture, (2) Low internal noise, (3) Large sensitive surface area per pixel. Since we can make pixels as big as we like with both CCD and CMOS, that third factor cancels out when comparing them. So let's look at (1) and (2).

(1) Sensitivity/Quantum Efficiency: by using microlenses to direct the light into the active areas of each pixel, CMOS are now ahead of front-illuminated CCDs and gaining on back-illuminated ones. Silicon is silicon in both cases; once photons are on target for an active pixel well, photons will be absorbed in depth distributions proportional to their wavelength in both cases.

(2) Internal noise: At normal (photographic) readout speeds, modern CMOS cameras are several times lower in readout noise than CCDs. Many are in the 2-electron area. S-CMOS is at 1 electron. CCD-based photographic cameras are at best around  6 or 7 electrons, and medium format digital's current best is 12 electrons.
Then there's dark noise. It is almost negligible in good CMOS sensors at everyday temperatures, but still woeful in uncooled CCDs (cooled means taking it to several 10's of degrees C below ambient temperature, as in an astronomical camera). A 300 second exposure with a modern Canon CMOS, without dark frame subtraction, looks as good as a 3 second exposure with many CCD cameras, also without dark frame subtraction. All medium format digital backs/cameras require long dark frame subtractions to alleviate this noise, but of course you cannot subtract its random (poisson) fluctuations, so that adds to the internal noise.

Ray
Title: Re: CCD and CMOS
Post by: jduncan on September 27, 2011, 07:58:13 AM
Your friend is wrong - or at least, quoting out of date information. CMOS has overtaken CCD in noise performance and is now beginning to displace CCDs in research instrumentation. I'm an astronomer too; my collaborators have just bought some new cameras for use in a high speed photometer on one of the world's largest telescopes - and guess what, they are CMOS (Andor S-CMOS), not CCD. Future space sensors will be CMOS too; just remember that space missions often take 10 years to develop, and usually deploy very old, reliable, tested-to-death, radiation-hardened technology, so it takes a while for new tech to work through to launch.

Excellent High ISO performance requires 3 things: (1) Efficient light capture, (2) Low internal noise, (3) Large sensitive surface area per pixel. Since we can make pixels as big as we like with both CCD and CMOS, that third factor cancels out when comparing them. So let's look at (1) and (2).

(1) Sensitivity/Quantum Efficiency: by using microlenses to direct the light into the active areas of each pixel, CMOS are now ahead of front-illuminated CCDs and gaining on back-illuminated ones. Silicon is silicon in both cases; once photons are on target for an active pixel well, photons will be absorbed in depth distributions proportional to their wavelength in both cases.

(2) Internal noise: At normal (photographic) readout speeds, modern CMOS cameras are several times lower in readout noise than CCDs. Many are in the 2-electron area. S-CMOS is at 1 electron. CCD-based photographic cameras are at best around  6 or 7 electrons, and medium format digital's current best is 12 electrons.
Then there's dark noise. It is almost negligible in good CMOS sensors at everyday temperatures, but still woeful in uncooled CCDs (cooled means taking it to several 10's of degrees C below ambient temperature, as in an astronomical camera). A 300 second exposure with a modern Canon CMOS, without dark frame subtraction, looks as good as a 3 second exposure with many CCD cameras, also without dark frame subtraction. All medium format digital backs/cameras require long dark frame subtractions to alleviate this noise, but of course you cannot subtract its random (poisson) fluctuations, so that adds to the internal noise.

Ray
Hi what about Infra RED? I am asking because I have this CCD DSLR and I was thinking of removing the IR filter.  From your argument the old adventages of CCD for visible spectrum no longer holds. Could you elaborate abut IR (for astro photo)

Thanks for your time,

J. Duncan
Title: Re: CCD and CMOS
Post by: fotometria gr on September 27, 2011, 07:55:46 PM
Could someone give me a technical explanation why CMOS can shoot at such high ISO's (nikon D700, etc) compared to CCD's despite CCD's being more sensitive to light and being a more expansive technology ? A friend of mine is a scientist and tells me that in astronomy and industrial applications the reverse of what is going on in the camera business is true, CCD's are more light sensitive and have less noise compared to CMOS. I noticed with most cameras that use CCD's (like the phase backs) anything over 800iso is not really considered to be great. The sensors that go to space are also CCD's.
All MFDBs are CCD (up to now), the P1 P25+ and P45+ are the world leaders in low light because of their ability to keep the sensor cool for long exposures, this is due to the great cooling system not the type of sensor used, the MFDBs are worst in their ISO performance because they lack any kind of noise reduction filter or circuit, this is a choice of the manufacturers due to the demand of the MF market for purity so that low ISO performance is maximum! The IQ of DBs is well beyond any modern DSLR even if they are 5 or 6 years old! The reason for CMOS sensors is mainly the low cost of production, but they do cost a lot more than CCDs to develop. I guess that the production volume of DSLRs compensates for the high development cost of CMOS. CCDs are considered sharper and with greater color than CMOS, I'm not aware which technology is better for high Iso, to compare that we should have a comparison of cameras with both technologies but the same NR system. Regards, Theodoros. www.fotometria.gr
Title: Re: CCD and CMOS
Post by: bjanes on September 27, 2011, 08:26:36 PM

Excellent High ISO performance requires 3 things: (1) Efficient light capture, (2) Low internal noise, (3) Large sensitive surface area per pixel. Since we can make pixels as big as we like with both CCD and CMOS, that third factor cancels out when comparing them. So let's look at (1) and (2).

(1) Sensitivity/Quantum Efficiency: by using microlenses to direct the light into the active areas of each pixel, CMOS are now ahead of front-illuminated CCDs and gaining on back-illuminated ones. Silicon is silicon in both cases; once photons are on target for an active pixel well, photons will be absorbed in depth distributions proportional to their wavelength in both cases.

Microlenses will collect light and focus it on the active sensing area of the sensel, but what about charge density, which limits the full well capacity of a sensel? The transistors in a CMOS will take up space and reduce the active sensing area of the sensel, leading to increased charge density in the CMOS as compared to CCD. I don't quite agree with point 3. For a given sensor size, pixel size is inversely proportional to the megapixel count. You can make the sensor larger, but at considerable expense.

Regards,

Bill
Title: Re: CCD and CMOS
Post by: ErikKaffehr on September 28, 2011, 12:32:11 AM
Hi Bill,

I sort of agree. I'm not really aware how large the physical fill factor is on CCD vs. CMOS, but the transistors definitively seem to take significant silicon area on CMOS. I presume that with narrower design rules the fill factor of CMOS is increasing. I'd also assume that FWC is proportional to physical fill factor.

Best regards
Erik


Microlenses will collect light and focus it on the active sensing area of the sensel, but what about charge density, which limits the full well capacity of a sensel? The transistors in a CMOS will take up space and reduce the active sensing area of the sensel, leading to increased charge density in the CMOS as compared to CCD. I don't quite agree with point 3. For a given sensor size, pixel size is inversely proportional to the megapixel count. You can make the sensor larger, but at considerable expense.

Regards,

Bill
Title: Re: CCD and CMOS
Post by: deejjjaaaa on September 28, 2011, 02:14:17 AM
CCDs are considered sharper and with greater color than CMOS

did you try to compare CCD-based P&S  w/ CMOS based FF dSLR  ;) ? so is the difference because of CCD vs CMOS or because of other factors  ::)
Title: Re: CCD and CMOS
Post by: deejjjaaaa on September 28, 2011, 02:17:07 AM
CCD-based photographic cameras are at best around  6 or 7 electrons, and medium format digital's current best is 12 electrons.

small CCD sensor based P&S like S95 are <= 4 electrons, are they not ?
Title: Re: CCD and CMOS
Post by: fotometria gr on September 28, 2011, 04:32:48 AM
did you try to compare CCD-based P&S  w/ CMOS based FF dSLR  ;) ? so is the difference because of CCD vs CMOS or because of other factors  ::)
1. No, neither I will.  :)  2. Don't know, is it? You seem to ignore "is considered" from my quote!  ;) Who cares anyway?  8) Regards, Theodoros. www.fotometria.gr
P.S. I prefer to judge products with what they do for my photography, don't care much to analyze the technical origin of the solution given from manufacturers nor the scientific theory behind it for that matter.
Title: Re: CCD and CMOS
Post by: ondebanks on September 28, 2011, 07:21:05 AM
Microlenses will collect light and focus it on the active sensing area of the sensel, but what about charge density, which limits the full well capacity of a sensel? The transistors in a CMOS will take up space and reduce the active sensing area of the sensel, leading to increased charge density in the CMOS as compared to CCD.

But Bill, we are talking about high-ISO performance - that was the original question. If you are forced into shooting in the high ISO regime, by definition you are not capturing much light/charge per pixel. So charge density and the full well capacity of the sensel are unconnected to this issue. These things matter when you have plenty of light per exposure, so you can shoot at low ISO, and can approach filling the FWC in the scene highlights.

I  don't quite agree with point 3. For a given sensor size, pixel size is inversely proportional to the megapixel count. You can make the sensor larger, but at considerable expense.

Point 3 was simply saying that if I can increase the pixel size/area while holding the readout noise constant, I gain more signal-to-noise per pixel, and get better high ISO. That, for example, is one of the factors which gave the Kodak DCS 620/720x DSLRs the edge in high ISO sports shooting 10 years ago - while maintaining the same 15 electrons readout noise as the 9 micron CCDs of the time, their CCDs had bigger 13 micron pixels, with twice the surface area, and twice the flux accumulated per pixel per unit exposure time. So bigger CCD pixels are better for high ISO. But, I can also make CMOS pixels bigger, and they gain S/N in the same way. And making pixels bigger reduces the megapixel count for both CCD and CMOS, also in the same way. So pixel size matters in absolute terms, but in this CCD vs. CMOS comparative question, this factor cancels out.

small CCD sensor based P&S like S95 are <= 4 electrons, are they not ?

Yes. I think this may be because there is a secondary dependence of readout noise on readout speed. P&S cameras like the S95 are not designed for fast full-resolution frame rates (the S95 reaches 0.9 fps), their users have no great expectations in that area, so the readout can be a little more "leisurely" in order to keep the readout noise down a bit. Astronomical CCD cameras often exploit this by taking the readout speed way down, sometimes to the kHz regime rather than the MHz regime. We are always working at the limit of the S/N, so when you've waited 1 hour for an exposure to complete, you don't really mind waiting another minute for it to be read out, if it keeps the noise down.

I wish that MFDB manufacturers would introduce this as a selectable feature on their backs - a slower, lower noise readout option, for those who can wait a few seconds longer between shots.

did you try to compare CCD-based P&S  w/ CMOS based FF dSLR  ;) ? so is the difference because of CCD vs CMOS or because of other factors  ::)

That comparison would be unfair, because of the vast differences in pixel size.

Hi what about Infra RED? I am asking because I have this CCD DSLR and I was thinking of removing the IR filter.  From your argument the old adventages of CCD for visible spectrum no longer holds. Could you elaborate abut IR (for astro photo)

Good news for you J.: for IR, in principle neither CMOS nor CCD should have any advantage over the other. But, even with the IR-block filter removed, it does depend an awful lot on the transmission curves of the Bayer RGB filters over the pixels. In fact, the differences between various CCD Bayer transmissions can be larger than the differences between popular CCD and CMOS Bayer transmissions. This is clear from manufacturer datasheets. In MF CCD sensors, for example, the Dalsa CCDs I've seen transmit IR predominantly through the Red-filtered pixels (that's 1 in 4 pixels). In Kodak MF CCDs, the Green- and Blue-filtered pixels tend to recover in transmission as the IR band begins beyond 700nm, and beyond about 820nm, every pixel is transmitting equally; allowing shorter IR exposures, and giving monochromatic B&W IR images which in principle require no de-Bayering interpolation, which would allow perfect per-pixel sharpness.

All MFDBs are CCD (up to now), the P1 P25+ and P45+ are the world leaders in low light because of their ability to keep the sensor cool for long exposures, this is due to the great cooling system not the type of sensor used, the MFDBs are worst in their ISO performance because they lack any kind of noise reduction filter or circuit, this is a choice of the manufacturers due to the demand of the MF market for purity so that low ISO performance is maximum! The IQ of DBs is well beyond any modern DSLR even if they are 5 or 6 years old! The reason for CMOS sensors is mainly the low cost of production, but they do cost a lot more than CCDs to develop. I guess that the production volume of DSLRs compensates for the high development cost of CMOS. CCDs are considered sharper and with greater color than CMOS, I'm not aware which technology is better for high Iso, to compare that we should have a comparison of cameras with both technologies but the same NR system. Regards, Theodoros. www.fotometria.gr

Theodoros, there's quite a bit of confusion and mythology in your response. I'll take each in turn:

"the P1 P25+ and P45+ are the world leaders in low light because of their ability to keep the sensor cool for long exposures" - Here you are confusing long-exposure capability with high-ISO capability. They depend on different things. That Kodak DCS 720x I mentioned above, for example (I got one for a song recently) - excellent at high ISO, dreadful at long exposures. Nice interchangeable viewfinders though!

the MFDBs are worst in their ISO performance because they lack any kind of noise reduction filter or circuit - It's vital that people understand that noise reduction trickery is not the reason why MFDBs cannot compete favourably with CMOS DSLRs on high ISO per-pixel performance. You can turn off the noise reduction completely, on say a Canon DSLR, and save a RAW file which has equal "purity" (to use your term). Apples with apples, the Canon still wins.

The IQ of DBs is well beyond any modern DSLR even if they are 5 or 6 years old! - At low ISO, yes. At high ISO, which is what we are talking about, no.

CCDs are considered sharper and with greater color than CMOS - Take the AA filter off a CMOS sensor and it is equally sharp as a CCD. Really, really. For that matter, put an AA filter over a CCD (I can, with my Kodak), and it is equally unsharp as a typical CMOS. As for color, that depends on the Bayer filters and the IR-blocking filter; there is no reason in principle, other than manufacturer design decisions on these filters, why the colour response of a CMOS unit cannot be the same as that of any given CCD.

What I'm saying is that you have to separate out what is intrinsic to the sensor (things like readout and dark noise) and things which are extrinsic (things like AA, IR-blocking, & Bayer filters layered on top of the sensor). Don't judge "CCD vs CMOS" on the basis of extrinsic things, which can be chopped and changed or easily redesigned. Kodak, for example, often offers the same intrinsic CCD sensor with different configurations of Bayer or no Bayer, IR-blocking coverglass or no coverglass.

I'm not aware which technology is better for high Iso, to compare that we should have a comparison of cameras with both technologies but the same NR system. - Or even better, with all NR turned off, and a RAW comparison with complete "purity". Well, that comparison has been done, over and over: it's an entire website called DxOMark! And I've done my own little version, comparing my MFDB to my Canon 5DII. Surprise, surprise, the Canon trounced the MFDB.  ;) I still prefer the MFDB at low ISO and short exposures, and the handling of a MF camera.

P.S. I prefer to judge products with what they do for my photography, don't care much to analyze the technical origin of the solution given from manufacturers nor the scientific theory behind it for that matter.

That's a perfectly valid position to take, but then why did you depart from it, and feel you had to get involved in this thread?  ???

Ray
Title: Re: CCD and CMOS
Post by: telyt on September 28, 2011, 08:26:16 AM
I wish that MFDB manufacturers would introduce this as a selectable feature on their backs - a slower, lower noise readout option, for those who can wait a few seconds longer between shots.

Leica is looking into this as a firmware update for the S2.  It would be a user-selectable menu option.
Title: Re: CCD and CMOS
Post by: TH_Alpa on September 28, 2011, 10:36:06 AM
Thanks Ray, for this detailled and interesting information.

Thierry

But Bill, we are talking about high-ISO performance ....
.....
Ray
Title: Re: CCD and CMOS
Post by: bjanes on September 28, 2011, 11:51:09 AM
small CCD sensor based P&S like S95 are <= 4 electrons, are they not ?

That is true, but a read noise of 4 electrons for a small pixel has a quite different effect on measured noise than for a large pixel because of camera gain, which is the number of electrons collected per data number (raw value). Since the P&S sensors use 12 bits per pixel for their output, I will use this bit depth for comparison.

A small pixel camera such as the Canon S70 may collect 8200 electrons and has a read noise of 3.2 electrons. Assuming that the raw pixel value at full well is near the maximum 12 bit pixel value of 4095, the gain is 2.0 electrons / data number. A larger pixel, such as with the Nikon D3, can collect about 66,500 electrons with a read noise of 4.9 electrons and a gain of 16 electrons /data number. The data are from Roger Clark (http://www.clarkvision.com/articles/digital.sensor.performance.summary/).

When we look at measured noise in the image, we are concerned with data numbers, not electrons. A read noise of 3.2 electrons for the S70 would translate to 3.2/2.0 or 1.6 data numbers. The read noise in terms of data numbers for the D3 would be 4.4/16 or 0.27 data numbers.

Regards,

Bill
Title: Re: CCD and CMOS
Post by: bjanes on September 28, 2011, 11:54:58 AM
Thanks Ray, for this detailled and interesting information.

Thierry


+1

Bill Janes
Title: Re: CCD and CMOS
Post by: theguywitha645d on September 28, 2011, 01:10:15 PM
In Kodak MF CCDs, the Green- and Blue-filtered pixels tend to recover in transmission as the IR band begins beyond 700nm, and beyond about 820nm, every pixel is transmitting equally; allowing shorter IR exposures, and giving monochromatic B&W IR images which in principle require no de-Bayering interpolation, which would allow perfect per-pixel sharpness.

Ray, this is very interesting. I believe my Pentax 645D has a Kodak sensor and I was thinking of trying some IR (but not remove the IR cut filter). Pentax has really worked on the performance of this sensor with ISOs of 1600 and no limit on integration time--I have done some wide-field astro images with exposures up to 5 minutes and known folks to shoot up to half an hour. You wouldn't know where I could get the spectral response of a Kodak 44x33mm 40MP Bayer sensor?
Title: Re: CCD and CMOS
Post by: hjulenissen on September 28, 2011, 02:56:12 PM
Where is the "cutting point" in exposure time when film generally give better results than digital?
Title: Re: CCD and CMOS
Post by: Radu Arama on September 28, 2011, 03:04:06 PM
Ray, this is very interesting. I believe my Pentax 645D has a Kodak sensor and I was thinking of trying some IR (but not remove the IR cut filter). Pentax has really worked on the performance of this sensor with ISOs of 1600 and no limit on integration time--I have done some wide-field astro images with exposures up to 5 minutes and known folks to shoot up to half an hour. You wouldn't know where I could get the spectral response of a Kodak 44x33mm 40MP Bayer sensor?

Hello,

I believe this is what you were looking for: http://www.kodak.com/ek/uploadedFiles/Content/Small_Business/Images_Sensor_Solutions/Datasheets(pdfs)/KAF-40000LongSpec.pdf the spectral response is on page 15.

Best regards,
Radu
Title: Re: CCD and CMOS
Post by: PierreVandevenne on September 28, 2011, 04:22:11 PM
Your friend is wrong - or at least, quoting out of date information. CMOS has overtaken CCD in noise performance and is now beginning to displace CCDs in research instrumentation. I'm an astronomer too; my collaborators have just bought some new cameras for use in a high speed photometer on one of the world's largest telescopes - and guess what, they are CMOS (Andor S-CMOS), not CCD.

I feel that answer is a bit misleading. Aren't the keywords "high speed" here? You can find all kinds of instruments in professional telescopes, including some based on relatively exotic designs (we aren't getting Aladdin III In:Sb sensors in our cameras any time soon, I think). CMOS based architectures are of course widely used in fields where high speed is possible or desirable (photometry of occultations for example, solar observation etc...). But CCD still reigns in imaging applications. I was so surprised by the above statement that I double checked what current major observatories use as imagers

ESO Paranal - http://www.eso.org/sci/facilities/paranal/instruments/index.html - have a look at the detailed description of the instruments, too many to list here

Gran Telescopio Canarias - http://www.gtc.iac.es/en/pages/instrumentation/osiris.php#Detector

Subaru - http://www.naoj.org/Observing/Instruments/SCam/

If the purpose is going deep and long exposures, everyone seems to still be using CCDs

Not that I disagree with the increased usefulness of CMOS based sensors in many fields in general - but do you have examples of CMOS sensors used for image acquisition in fairly long exposures?



Title: Re: CCD and CMOS
Post by: theguywitha645d on September 28, 2011, 04:59:12 PM
Hello,

I believe this is what you were looking for: http://www.kodak.com/ek/uploadedFiles/Content/Small_Business/Images_Sensor_Solutions/Datasheets(pdfs)/KAF-40000LongSpec.pdf the spectral response is on page 15.

Best regards,
Radu

Radu, thank you. That is very interesting...
Title: Re: CCD and CMOS
Post by: fotometria gr on September 28, 2011, 05:36:12 PM
  Ray,
1. On your first quote on me, I suggest that perhaps there was ....low oxygen in the place where you read my post!  ???  :o What are you talking about? It clearly has nothing to do with my statement!  :-X  8)
2. On your second quote on me there was certainly much CO2 present in your room!  ;D  ;) Its just an answer to the quoter not a post on the OP!  :'(  :-*
3. Neither yours nor any other Canon turn their NR off when you instruct them to do so from menu, it stays on to some extend by default!  :P Cheers, Theodoros www.fotometria.gr
Title: Re: CCD and CMOS
Post by: bjanes on September 28, 2011, 05:55:06 PM
That is true, but a read noise of 4 electrons for a small pixel has a quite different effect on measured noise than for a large pixel because of camera gain, which is the number of electrons collected per data number (raw value). Since the P&S sensors use 12 bits per pixel for their output, I will use this bit depth for comparison.

A small pixel camera such as the Canon S70 may collect 8200 electrons and has a read noise of 3.2 electrons. Assuming that the raw pixel value at full well is near the maximum 12 bit pixel value of 4095, the gain is 2.0 electrons / data number. A larger pixel, such as with the Nikon D3, can collect about 66,500 electrons with a read noise of 4.9 electrons and a gain of 16 electrons /data number. The data are from Roger Clark (http://www.clarkvision.com/articles/digital.sensor.performance.summary/).

When we look at measured noise in the image, we are concerned with data numbers, not electrons. A read noise of 3.2 electrons for the S70 would translate to 3.2/2.0 or 1.6 data numbers. The read noise in terms of data numbers for the D3 would be 4.4/16 or 0.27 data numbers.

On rechecking my figures, I see that the read noise given by Roger Clark applies only to higher ISOs and can not be achieved by the D3 at base ISO because of noise in the electronics downstream to the sensor. Peter Facey (http://www.brisk.org.uk/photog/d3gainiso200.html) has performed a detailed analysis of the D3 and the read noise at ISO 200 (the base ISO) is 4.5 14 bit DN and the gain is 4.13 electrons/DN. The read noise at ISO 200 is therefore 18.36 electrons. The 12 bit gain would be 16.5 e-/DN, and the noise in 12 bit DN would be 1.1 DN. At base ISO the D3 is not much better than a P&S in terms of read noise, but the S:N would be much higher since more signal was collected. Sorry for the confusion.  :(

On newer CMOS sensors such as the Sony used by the D7000, read noise does not vary with ISO. For more details on the topic of sensor vs. camera performance, see Emil Martinec (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3a.html).

Regards,

Bill
Title: Re: CCD and CMOS
Post by: theguywitha645d on September 28, 2011, 06:01:00 PM
Where is the "cutting point" in exposure time when film generally give better results than digital?
That is a lot more complicated as film suffers from reciprocity law failure. So with a long exposure, you will not get noise with film, but you will not be capturing as many photons (or rather the photosites on the grain lose energy and unexpose, so to speak). If you have one stop of compensation with a calculated 15 minute exposure, then a 15 minute digital exposure would need to be compared to a 30 minute film exposure. The digital camera will not have a change in contrast either. At some point, film will give up completely and not be able to record information whereas a sensor can still be counting photons. I guess a film exposure can be much better in very hot environments. A digital back would need active cooling and as far as I know, only scientific cameras do that.
Title: Re: CCD and CMOS
Post by: ErikKaffehr on September 29, 2011, 01:45:58 AM
Hi,

Since it seems that you have done some practical tests, and real world comparisons between MF and DSLRs are rare, it would be nice if you could share your findings.

- Do your findings match DxO-mark data?
- In what area "trounces" Canon MFDB?
- Obviously you prefer MFDB to Canon in low ISO. What benefits do you see?
- The benefits that you see, can they be explained by sensor size (collecting more Photons, higher MTF for given feature size and better edge contrast due to lack of OLP filtering, etc) or do they involve "magic qualities"?
- If you see "magic qualities" can they be explained?

Best regards
Erik


Quote
And I've done my own little version, comparing my MFDB to my Canon 5DII. Surprise, surprise, the Canon trounced the MFDB.   I still prefer the MFDB at low ISO and short exposures, and the handling of a MF camera.

Title: Re: CCD and CMOS
Post by: ErikKaffehr on September 29, 2011, 01:49:52 AM
Hi,

Nikon D3X has a similar option, it can use either 12-bit or 14-bit output. The 14-bit ouput is much slower. It is not known how 14-bits are achieved, but could be longer integration times.

Best regards
Erik


Leica is looking into this as a firmware update for the S2.  It would be a user-selectable menu option.
Title: Re: CCD and CMOS
Post by: hjulenissen on September 29, 2011, 06:59:47 AM
Neither yours nor any other Canon turn their NR off when you instruct them to do so from menu, it stays on to some extend by default! 
Lots of hot air, lots of opinion, no references.

-h
Title: Re: CCD and CMOS
Post by: fotometria gr on September 29, 2011, 07:43:47 AM
Lots of hot air, lots of opinion, no references.

-h
Its true that temprature can rise up if people distort completely what you state and call you "confused", as for references, if my British MechEng BEng degree doesn't do, have a look at DPreview on any  modern Dslr, they state the same. Regards, Theodoros. www.fotometria.gr
Title: Re: CCD and CMOS
Post by: ondebanks on September 29, 2011, 08:39:26 AM
I feel that answer is a bit misleading. Aren't the keywords "high speed" here? You can find all kinds of instruments in professional telescopes, including some based on relatively exotic designs (we aren't getting Aladdin III In:Sb sensors in our cameras any time soon, I think). CMOS based architectures are of course widely used in fields where high speed is possible or desirable (photometry of occultations for example, solar observation etc...). But CCD still reigns in imaging applications. I was so surprised by the above statement that I double checked what current major observatories use as imagers

ESO Paranal - http://www.eso.org/sci/facilities/paranal/instruments/index.html - have a look at the detailed description of the instruments, too many to list here

Gran Telescopio Canarias - http://www.gtc.iac.es/en/pages/instrumentation/osiris.php#Detector

Subaru - http://www.naoj.org/Observing/Instruments/SCam/

If the purpose is going deep and long exposures, everyone seems to still be using CCDs

Not that I disagree with the increased usefulness of CMOS based sensors in many fields in general - but do you have examples of CMOS sensors used for image acquisition in fairly long exposures?


Pierre,

What I said was "CMOS...is now beginning to displace CCDs in research instrumentation". Not "CMOS has displaced". The instruments you linked to are all relatively old; some were completed in the 1990s, the rest commenced development before the recent surge in CMOS performance. They generate science, not income, so they won't be replaced unless absolutely necessary. It's going to take time.

Also, there's a certain amount of inertia imposed by the different readout electronics involved with CMOS. Observatories like to standardize on and re-use existing CCD controller systems as far as possible, for multiple instruments or when upgrading a sensor in an instrument (google ULTRADAS and SDSU-II for example). Switching to CMOS will require a new system, more development time, new documentation, new training, and above all more funding, which is really hard to get.

Another reason is that we still await large low-noise CMOS sensors (the same reason why there are still no CMOS MF digital backs). Large research telescopes have giant focal planes to populate with mosaics of imaging sensors, so the availability of large CCDs keeps them at the forefront.

You are right insofar as CMOS usage in research at present is mainly in the high-speed or timeseries niche. Any scenario where you have to take many frames, is where their low readout noise per frame makes the most difference. Their other great advantage, low dark noise, is a bit moot in research where liquid nitrogen cryostats are normally used to take CCD dark current down to an acceptably low level. But modest (peltier based) cooling on CMOS is around the same level (google CentralDS for example), and I can imagine that moving away from the hassle and expense of cryogenics (not just in plant and materials, but also because technicians must be employed to refill dewars 2 or 3 times in every 24 hours period - the sensor must never be allowed to warm up) will be very attractive to observatories.

You ask if I "have examples of CMOS sensors used for image acquisition in fairly long exposures"? I presume you mean discounting amateur astro-imagers, who use both off-the-shelf and modified CMOS DSLRs for exposures running to hours net? Well in research there are areas like wide-field auroral monitoring which use similar setups.

Ray
Title: Re: CCD and CMOS
Post by: BartvanderWolf on September 29, 2011, 09:40:57 AM
Neither yours nor any other Canon turn their NR off when you instruct them to do so from menu, it stays on to some extend by default!

I've never seen any proof for that assumption, I did see proof to the opposite. A simple method to detect noise reduction is to shoot a well focused image of a "White noise" patch, and then display the Fourier transform. The gradual signal decline towards higher spatial frequencies (caused by MTF, finite sensel size, and AA-filter for suppressing Bayer CFA related aliasing) is pretty gradual, unlike the Fourier transform of a noise reduced image. The Fourier transform of the Read noise alone also doesn't show signs of noise reduction.

If you have any trustworthy proof for your assertion, a lot of people would be interested, I'm sure.

Cheers,
Bart    
Title: Re: CCD and CMOS
Post by: eronald on September 29, 2011, 10:15:27 AM
Link to such a patch?

I've never seen any proof for that assumption, I did see proof to the opposite. A simple method to detect noise reduction is to shoot a well focused image of a "White noise" patch, and then display the Fourrier transform. The gradual signal decline towards higher spatial frequencies (caused by MTF, finite sensel size, and AA-filter for suppressing Bayer CFA related aliasing) is pretty gradual, unlike the Fourier transform of a noise reduced image. The Fourier transform of the Read noise alone also doesn't show signs of noise reduction.

If you have any trustworthy proof for your assertion, a lot of people would be interested, I'm sure.

Cheers,
Bart   
Title: Re: CCD and CMOS
Post by: PierreVandevenne on September 29, 2011, 10:37:33 AM
What I said was "CMOS...is now beginning to displace CCDs in research instrumentation". Not "CMOS has displaced". The instruments you linked to are all relatively old; some were completed in the 1990s, the rest commenced development before the recent surge in CMOS performance. They generate science, not income, so they won't be replaced unless absolutely necessary. It's going to take time.
Another reason is that we still await large low-noise CMOS sensors (the same reason why there are still no CMOS MF digital backs). Large research telescopes have giant focal planes to populate with mosaics of imaging sensors, so the availability of large CCDs keeps them at the forefront.
You are right insofar as CMOS usage in research at present is mainly in the high-speed or timeseries niche. Any scenario where you have to take many frames, is where their low readout noise per frame makes the
You ask if I "have examples of CMOS sensors used for image acquisition in fairly long exposures"? I presume you mean discounting amateur astro-imagers, who use both off-the-shelf and modified CMOS DSLRs for exposures running to hours net? Well in research there are areas like wide-field auroral monitoring which use similar setups.

On the age of instrumentation aspect, that's the reason I included the Gran Canaria Telescope - 2009. I could have included SALT as well (http://www.salt.ac.za/telescope/instrumentation/salticam/) which is being finished right now. And yes, I was discounying DSLR amateur exposures (although I have done quite a few of them) as they aren't really indicative of what's the cutting edge in real scientific instruments is.
Anyway I don't want to turn this in a long argument because I am sure we agree on the fundamentals anyway.
Title: Re: CCD and CMOS
Post by: ondebanks on September 29, 2011, 01:03:32 PM
 Ray,
1. On your first quote on me, I suggest that perhaps there was ....low oxygen in the place where you read my post!  ???  :o What are you talking about? It clearly has nothing to do with my statement!  :-X  8)
2. On your second quote on me there was certainly much CO2 present in your room!  ;D  ;) Its just an answer to the quoter not a post on the OP!  :'(  :-*
3. Neither yours nor any other Canon turn their NR off when you instruct them to do so from menu, it stays on to some extend by default!  :P Cheers, Theodoros www.fotometria.gr

Theodoros, clearly you are a man of many, and very strong, emotions. 1 short message, 9 emoticons!

Could you please identify the offending parts of my "first" and "second" quotes to you? As I am having trouble matching up your objections to what I had said.

On your 3rd point, if what you say were true, no-one would be able to produce a photon-transfer curve which obeys Poisson statistics, from Canon RAW files. So, how do you explain the fact that several Canon users have successfully done so?

When you say that Canon's NR cannot be turned off, are you perhaps thinking of Canon's on-chip double-correlated sampling? That is not noise reduction in the sense that everyone assumes; it is better described as noise prevention. In no way does it usurp the integrity, statistics, or spatial correlation of the signal, unlike true "NR". It's kosher. This  (http://www.luminous-landscape.com/forum/index.php?topic=31741.0)is a good LL thread on that topic (page 2 is the important part).

Ray
Title: Re: CCD and CMOS
Post by: BartvanderWolf on September 29, 2011, 03:01:25 PM
Link to such a patch?

Hi Edmund,

Totally random noise will do, even Photoshop could produce something usable.
A program like ImageJ has plugins to create various types of noise.

http://nanophotonics.ece.cornell.edu/Publications/randomTargetMethod.pdf (http://nanophotonics.ece.cornell.edu/Publications/randomTargetMethod.pdf)

Cheers,
Bart

Title: Re: CCD and CMOS
Post by: fotometria gr on September 29, 2011, 03:24:32 PM
Theodoros, clearly you are a man of many, and very strong, emotions. 1 short message, 9 emoticons!

Could you please identify the offending parts of my "first" and "second" quotes to you? As I am having trouble matching up your objections to what I had said.

On your 3rd point, if what you say were true, no-one would be able to produce a photon-transfer curve which obeys Poisson statistics, from Canon RAW files. So, how do you explain the fact that several Canon users have successfully done so?

When you say that Canon's NR cannot be turned off, are you perhaps thinking of Canon's on-chip double-correlated sampling? That is not noise reduction in the sense that everyone assumes; it is better described as noise prevention. In no way does it usurp the integrity, statistics, or spatial correlation of the signal, unlike true "NR". It's kosher. This  (http://www.luminous-landscape.com/forum/index.php?topic=31741.0)is a good LL thread on that topic (page 2 is the important part).

Ray

1. You clearly quoted that I was confusing LONG EXPOSURE with HIGH ISO performance, while you are talking to a 30 years "quite well" known photographer that is at least respected for his work at more than one countries as if his knowledge was the knowledge of an ignorant kid! Clearly to take a part of the whole statement to "analyze" like if the the answer wasn't in front of you on the rest of the statement, is because: a) you intented to do so ....or b) there was not much oxygen in the room and you missed it from "eye blaring"   :-[
2. Your second INSULT, has been answered already  :o  ???
3. It has been answered already to another quoter and you PARTIALLY accepted it up there, ......I've no intention (although I can) to be drugged in such a conversation, There is NO DSLR that turns its NR OFF when its instructed by the user to do so and .....thats it!! If you have a proof for the opposite (PROOF NOT THEORY like ...if it was so... then.... so it should.... that article says....), then ....I can interfere for you to have a really good job in the industry, ......I will even help you in your photography if you want!  ::) Regards, Theodoros. www.fotometria.gr
Title: Re: CCD and CMOS
Post by: bjanes on September 29, 2011, 03:58:52 PM
Totally random noise will do, even Photoshop could produce something usable.
A program like ImageJ has plugins to create various types of noise.

Shot noise is more or less random and one can test for filtering of the image of a raw file. Shown below is an image of the green 1 channel of a raw file from the Nikon D3 and the same raw file rendered in ACR, which involves noise reduction. Fourier analysis was done with ImageJ. The D3 does not filter the raw image. Readers could repeat the test with their own cameras. One can use Iris to separate the raw channels.

(http://bjanes.smugmug.com/Photography/Raw/i-VgBhwzc/0/O/fft123composite.png)

Regards,

Bill
Title: Re: CCD and CMOS
Post by: FranciscoDisilvestro on September 29, 2011, 04:21:49 PM
According to this study (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p4.html) by Emil Martinec, The D3 and D300 apply noise reduction at exposures of 1/4 sec. and longer which can't be turned off by the user.
Title: Re: CCD and CMOS
Post by: fotometria gr on September 29, 2011, 04:31:30 PM
I've never seen any proof for that assumption, I did see proof to the opposite. A simple method to detect noise reduction is to shoot a well focused image of a "White noise" patch, and then display the Fourrier transform. The gradual signal decline towards higher spatial frequencies (caused by MTF, finite sensel size, and AA-filter for suppressing Bayer CFA related aliasing) is pretty gradual, unlike the Fourier transform of a noise reduced image. The Fourier transform of the Read noise alone also doesn't show signs of noise reduction.

If you have any trustworthy proof for your assertion, a lot of people would be interested, I'm sure.

Cheers,
Bart   
It has been answered to a previous quoter Bart, I'm sorry I don't intend, want, or am in a mood to spend pages (that is what is needed) to prove it. Although I have a British BEng in MechEng, I am a photographer and thus I find such conversations time consuming and useless, (its also that I haven't really practice my English for more than 20 years and thus its even more painful and time consuming for me). It is so though... all Dslrs have NR by default and if you happen to have some friends that work in the manufacturing of MFDBs or in sensor constructing (like I do), they would verify this to you! Please look at it..., you'll find that I'm right! Dslrs are constructed for the "enthusiast" or "advanced" user, (industry terms) which is considered by manufacturers to be a ....(dragged by the nose junkie) that will never notice the difference! This is exactly the reason why photo equipment "tests" are done by "experts" that don't have even one photograph or work published and why "sample photos" of these tests is bridges, or castles, or beaches, or flowers..... thats what the industry thinks of "target group"! Regards, Theodoros. www.fotometria.gr
Title: Re: CCD and CMOS
Post by: hjulenissen on September 29, 2011, 04:42:31 PM
as for references, if my British MechEng BEng degree doesn't do,
Sorry, does not impress me. If that education was worth anything, surely they must have taught you what people mean when they ask for references?

-h
Title: Re: CCD and CMOS
Post by: fotometria gr on September 29, 2011, 05:27:13 PM
Sorry, does not impress me. If that education was worth anything, surely they must have taught you what people mean when they ask for references?

-h
Who gives a ....whatever if you was impressed or not?  ??? DSLRs still don't turn their NR off!  :P ....and what you are doing by erasing the rest of the sentence is at least .....immoral!  :-* You obviously did so to object people of reading it, although it was only a few more words...  ???  >:(  ;D  :P Cheers (well..... not really) Theodoros.  8) www.fotometria.gr
Title: Re: CCD and CMOS
Post by: ondebanks on September 29, 2011, 05:45:58 PM
According to this study (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p4.html) by Emil Martinec, The D3 and D300 apply noise reduction at exposures of 1/4 sec. and longer which can't be turned off by the user.

Yup, Nikon are bad boys, but only for long exposures. If you search my posting history, I've often chastised Nikon for this. And also, for the way they subtract a bias level from the RAW, which makes it harder (but not impossible) to measure the readnoise, and impossible to fully subtract bias pattern noise from images like flatfields, which is of course nasty when one then comes to divide by a flatfield...as someone has noted here before, it's mostly the astrophotographers who moan about that!

It's nicely explained by Jerry Lodrigus here (http://www.astropix.com/HTML/I_ASTROP/NIK_CAN.HTM); scroll down to points 5, 6 and 7 under the "Nikon" heading.

Canon, emphatically, do not alter their RAW files in long exposures - or short ones, for that matter. Nor do they subtract the bias level. RAW means RAW with them.
Title: Re: CCD and CMOS
Post by: FranciscoDisilvestro on September 29, 2011, 06:09:03 PM
Very interesting, so, according to another link referenced in the Jerry Lodrigus page, the way to avoid the noise reduction for long exposures from Nikon cameras is to turn on Long Exposure Noise Reduction and turn the camera off during the dark frame acquisition. I wouldn't have attempted to do that.
Title: Re: CCD and CMOS
Post by: ondebanks on September 29, 2011, 06:42:26 PM
1. You clearly quoted that I was confusing LONG EXPOSURE with HIGH ISO performance, while you are talking to a 30 years "quite well" known photographer that is at least respected for his work at more than one countries as if his knowledge was the knowledge of an ignorant kid! Clearly to take a part of the whole statement to "analyze" like if the the answer wasn't in front of you on the rest of the statement, is because: a) you intented to do so ....or b) there was not much oxygen in the room and you missed it from "eye blaring"   :-[

Here is the entire, unedited first post you made in this thread. I encourage others to read it:

Quote from: larkis on September 25, 2011, 08:46:11 AM
Could someone give me a technical explanation why CMOS can shoot at such high ISO's (nikon D700, etc) compared to CCD's despite CCD's being more sensitive to light and being a more expansive technology ? A friend of mine is a scientist and tells me that in astronomy and industrial applications the reverse of what is going on in the camera business is true, CCD's are more light sensitive and have less noise compared to CMOS. I noticed with most cameras that use CCD's (like the phase backs) anything over 800iso is not really considered to be great. The sensors that go to space are also CCD's.


All MFDBs are CCD (up to now), the P1 P25+ and P45+ are the world leaders in low light because of their ability to keep the sensor cool for long exposures, this is due to the great cooling system not the type of sensor used, the MFDBs are worst in their ISO performance because they lack any kind of noise reduction filter or circuit, this is a choice of the manufacturers due to the demand of the MF market for purity so that low ISO performance is maximum! The IQ of DBs is well beyond any modern DSLR even if they are 5 or 6 years old! The reason for CMOS sensors is mainly the low cost of production, but they do cost a lot more than CCDs to develop. I guess that the production volume of DSLRs compensates for the high development cost of CMOS. CCDs are considered sharper and with greater color than CMOS, I'm not aware which technology is better for high Iso, to compare that we should have a comparison of cameras with both technologies but the same NR system. Regards, Theodoros. www.fotometria.gr

So - in answer to the OP's question, which asked only about high ISO and never once mentioned long exposures, your first explanation connects "low light" with "long exposures". What else were we to think, but that you were confused between them? We can only read it the way you write it.

2. Your second INSULT, has been answered already  :o  ???

I still have no idea what my insult was? If an apology is warranted, how can I apologise if I don't know what I'm apologising for?!


3. It has been answered already to another quoter and you PARTIALLY accepted it up there, ......I've no intention (although I can) to be drugged in such a conversation, There is NO DSLR that turns its NR OFF when its instructed by the user to do so and .....thats it!! If you have a proof for the opposite (PROOF NOT THEORY like ...if it was so... then.... so it should.... that article says....), then ....I can interfere for you to have a really good job in the industry, ......I will even help you in your photography if you want!  ::) Regards, Theodoros. www.fotometria.gr

Thanks, but I already have a really good job, as a university lecturer in physics and astronomy. I teach a course on signal processing, another on astronomical data analysis, and another on observational astronomy, with a particular emphasis on detector issues. I've been doing digital image processing since 1992; my PhD thesis concerned calibrating astronomical imaging photon-counting detectors called MAMAs, and their use in high-resolution imaging, photometry and deconvolution of the centres of star clusters.

But I guess a British MechEng BEng degree trumps that.

Now I AM annoyed, because you're the first person ever on LuLa to force me to play the "my credentials are better than yours" card, and I really hate doing that. The letters after one's name should have nothing to do with anything in a discussion like this. Only the ability to present proper evidence matters. We're still waiting for yours...
Title: Re: CCD and CMOS
Post by: PierreVandevenne on September 29, 2011, 07:29:11 PM
CCDs are considered sharper and with greater color than CMOS

CCD or CMOS's don't have "greater color". They don't have any color. They count photons. You can filter photons before they reach the CCD, by means of an individual filters or a filter matrix (for example Bayer).
Title: Re: CCD and CMOS
Post by: fotometria gr on September 29, 2011, 08:03:10 PM
Here is the entire, unedited first post you made in this thread. I encourage others to read it:

Quote from: larkis on September 25, 2011, 08:46:11 AM
Could someone give me a technical explanation why CMOS can shoot at such high ISO's (nikon D700, etc) compared to CCD's despite CCD's being more sensitive to light and being a more expansive technology ? A friend of mine is a scientist and tells me that in astronomy and industrial applications the reverse of what is going on in the camera business is true, CCD's are more light sensitive and have less noise compared to CMOS. I noticed with most cameras that use CCD's (like the phase backs) anything over 800iso is not really considered to be great. The sensors that go to space are also CCD's.


All MFDBs are CCD (up to now), the P1 P25+ and P45+ are the world leaders in low light because of their ability to keep the sensor cool for long exposures, this is due to the great cooling system not the type of sensor used, the MFDBs are worst in their ISO performance because they lack any kind of noise reduction filter or circuit, this is a choice of the manufacturers due to the demand of the MF market for purity so that low ISO performance is maximum! The IQ of DBs is well beyond any modern DSLR even if they are 5 or 6 years old! The reason for CMOS sensors is mainly the low cost of production, but they do cost a lot more than CCDs to develop. I guess that the production volume of DSLRs compensates for the high development cost of CMOS. CCDs are considered sharper and with greater color than CMOS, I'm not aware which technology is better for high Iso, to compare that we should have a comparison of cameras with both technologies but the same NR system. Regards, Theodoros. www.fotometria.gr

So - in answer to the OP's question, which asked only about high ISO and never once mentioned long exposures, your first explanation connects "low light" with "long exposures". What else were we to think, but that you were confused between them? We can only read it the way you write it.

I still have no idea what my insult was? If an apology is warranted, how can I apologise if I don't know what I'm apologising for?!


Thanks, but I already have a really good job, as a university lecturer in physics and astronomy. I teach a course on signal processing, another on astronomical data analysis, and another on observational astronomy, with a particular emphasis on detector issues. I've been doing digital image processing since 1992; my PhD thesis concerned calibrating astronomical imaging photon-counting detectors called MAMAs, and their use in high-resolution imaging, photometry and deconvolution of the centres of star clusters.

But I guess a British MechEng BEng degree trumps that.

Now I AM annoyed, because you're the first person ever on LuLa to force me to play the "my credentials are better than yours" card, and I really hate doing that. The letters after one's name should have nothing to do with anything in a discussion like this. Only the ability to present proper evidence matters. We're still waiting for yours...

OK! READ IT THE WAY I RIGHT IT!!!!  :o I suggest you need some therapy  ;D Excessive CO2 has caused you some very serious brain damage.  ;)  :)  :D  ;D Its really a pity for a PHD professor to humiliate himself by his own statements!  :o  ::) There is no person in the world "buddy" that would come in the assumptions that you came by reading what you have up there in bold WHICH IS MY STATEMENT AND THERE IS NOTHING WRONG WITH IT!!  :-* And please don't try to give me any advice of morality, SINCE ITS YOU THAT "JUMPED" ME WITHOUT ANY REASON WHATSOEVER!!  ???  :P I suggest you stop humiliating yourself and do some photography or read some Aristoteles (The greatest of philosophers and inventor of logic as a science)  ;)  :) AH! and open up the window..., fresh air will do you some good....  8) By the way... there is no DSLR (certainly no Canon) that turns completely its NR off when its instructed to do so from menu  :-* .....and there is nothing to change that no matter what you think or not.  :D ....and please don't bother me, answer to me, talk to me, refer to me, or think about me anymore!  ;D  :-* I don't even want your apology for stating that what you have up there in bold is a result of a confused mind! You don't exist for me!  :)  8) www.fotometria.gr
Title: Re: CCD and CMOS
Post by: fotometria gr on September 29, 2011, 08:20:38 PM
CCD or CMOS's don't have "greater color". They don't have any color. They count photons. You can filter photons before they reach the CCD, by means of an individual filters or a filter matrix (for example Bayer).

True enough, sensors are really B&W, I guess "considered" is the key word.... :)  I'm surprised how you confused "considered" (which refers to the result) with "have",  :o ....its really easy to goof between the two words i guess,  ::)  ....they are so near in grammar and vocabulary that can happen to anybody  :-X  8) Regards, Theodoros. www.fotometria.gr

Title: Re: CCD and CMOS
Post by: billthecat on September 29, 2011, 08:22:33 PM
I'd think that CCDs have better color since the value of each pixel is more matched to other pixels due to the different readout of CCD vs CMOS.

Bill

CCD or CMOS's don't have "greater color". They don't have any color. They count photons. You can filter photons before they reach the CCD, by means of an individual filters or a filter matrix (for example Bayer).

Title: Re: CCD and CMOS
Post by: madmanchan on September 29, 2011, 08:53:19 PM
I'd think that CCDs have better color since the value of each pixel is more matched to other pixels due to the different readout of CCD vs CMOS.

I don't follow you at all.  Can you please clarify what you mean by this?
Title: Re: CCD and CMOS
Post by: ErikKaffehr on September 29, 2011, 09:01:25 PM
No, that makes no sense at all. But it is quite possible that Phase, Leaf or Hasselblad would have a color grid array with transmission characteristic better matched to studio photography at low ISO compared to say Canon who may have a CGA more optimized at available light photography at high ISO. The same situation may apply to DSLR vendors, too. It is probable that Canon, Nikon and Sony have different CGA charactristics.

Best regards
Erik


I'd think that CCDs have better color since the value of each pixel is more matched to other pixels due to the different readout of CCD vs CMOS.

Bill

Title: Re: CCD and CMOS
Post by: jeremypayne on September 29, 2011, 09:51:53 PM
OK!

I really think you ought to find somewhere else to hang out on the internet.
Title: Re: CCD and CMOS
Post by: Schewe on September 29, 2011, 10:23:45 PM
OK! READ IT THE WAY I WRIGHT IT!!!!  :o

So, English isn't your first language? Did you mean "right it"?

Quote
P.S. Does anybody remember the story of the tower of Babel from the bible? I wonder why communication is falling lower and even lower by the day....

Uh...before you get all bent out of shape, maybe you should try multiple translations to see if YOU'VE made an error in translation?

I've read the thread and simply don't see why you are bent out of shape...try less caffeine or maybe change your meds?

We're having a discussion here, no need to get your panties in a bunch, right?
Title: Re: CCD and CMOS
Post by: bjanes on September 29, 2011, 10:33:43 PM
Yup, Nikon are bad boys, but only for long exposures. If you search my posting history, I've often chastised Nikon for this. And also, for the way they subtract a bias level from the RAW, which makes it harder (but not impossible) to measure the readnoise, and impossible to fully subtract bias pattern noise from images like flatfields, which is of course nasty when one then comes to divide by a flatfield...as someone has noted here before, it's mostly the astrophotographers who moan about that!

It's nicely explained by Jerry Lodrigus here (http://www.astropix.com/HTML/I_ASTROP/NIK_CAN.HTM); scroll down to points 5, 6 and 7 under the "Nikon" heading.

Canon, emphatically, do not alter their RAW files in long exposures - or short ones, for that matter. Nor do they subtract the bias level. RAW means RAW with them.


Emil and Jerry may be very knowledgeable and careful workers, but a FFT analysis of a 30 second Darkframe from the D3 is shown below and I see no evidence of filtering. Processing was as before: split_cfa in Iris and cropping of the green 1 channel to 300 x 300 px in Iris and analysis by Image J. All NR and dark frame subtraction was turned off. I can supply the raw file and the .fit files to anyone who wants to do their own critique.

(http://bjanes.smugmug.com/Photography/Raw/i-wPh9VPm/0/O/fftdarkframe.png)

Here is the FFT of the same file with the median filter applied with a radius of 4 showing clear cut evidence of filtering:

(http://bjanes.smugmug.com/Photography/Raw/i-LfChTTv/0/O/FFT-of-f1cropbmedian4fft.png)

And here is a PNG of the cropped green 1 channel

(http://bjanes.smugmug.com/Photography/Raw/i-M2zk7LM/0/O/f1cropb.png)

Regards,

Bill

Title: Re: CCD and CMOS
Post by: BartvanderWolf on September 30, 2011, 04:37:56 AM
Emil and Jerry may be very knowledgeable and careful workers, but a FFT analysis of a 30 second Darkframe from the D3 is shown below and I see no evidence of filtering.

Hi Bill,

I don't see any evidence of filtering in you sample either. Since I don't have a Nikon I cannot test it, but could it be that higher ISO settings behave different from lower ISOs?

Cheers,
Bart
Title: Re: CCD and CMOS
Post by: ondebanks on September 30, 2011, 05:09:44 AM
Emil and Jerry may be very knowledgeable and careful workers, but a FFT analysis of a 30 second Darkframe from the D3 is shown below and I see no evidence of filtering. Processing was as before: split_cfa in Iris and cropping of the green 1 channel to 300 x 300 px in Iris and analysis by Image J. All NR and dark frame subtraction was turned off. I can supply the raw file and the .fit files to anyone who wants to do their own critique.

Here is the FFT of the same file with the median filter applied with a radius of 4 showing clear cut evidence of filtering:

And here is a PNG of the cropped green 1 channel

Regards,

Bill


Nice work, Bill. Maybe Bart's right and Nikon set it up so that it depends on ISO, or else I guess Nikon must have changed their D3 firmware for the better since Christian Buil did his tests in 2007 which showed the D3 filtering (http://www.astropix.com/HTML/I_ASTROP/NIK_CAN.HTM) and similarly Bill Claff's tests shown by Emil Martinec for the D3, D300, and D200 (http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p4.html). If that's the case, then I applaud Nikon! If anyone else has a D3 or indeed any modern Nikon DSLR, I'd love to see if that can be confirmed. And Bill, if you have a chance could you take the exposure time out to say 5 or 10 minutes, and see if it is still ok? And also try at different ISO settings? Thanks.

Ray
Title: Re: CCD and CMOS
Post by: Fine_Art on September 30, 2011, 05:20:47 AM
I'd think that CCDs have better color since the value of each pixel is more matched to other pixels due to the different readout of CCD vs CMOS.

Bill


I think the argument is based on the camera testing at the end of manufacture. It is going to be tested against known images of blue , green, red, for output accuracy. The firmware is set to correct any deviation in the signal. In CMOS the correction is against each pixel. In CCD the conversion of photon to current is against whole rows so the converter is calibrated against 1000s of pixels output. Of course each pixel still has to output the same value so the correction of the color is the same. The accuracy of the signal converter should be much greater in CCD.

Title: Re: CCD and CMOS
Post by: PierreVandevenne on September 30, 2011, 07:36:17 AM
True enough, sensors are really B&W, I guess "considered" is the key word.... :)  I'm surprised how you confused "considered" (which refers to the result) with "have"

I didn't. But I simply find the foundations upon which you base your many claims to be a bit shaky, if only from a rethorical point of view. "is widely considered" is a sociological assessment rather than sound formal logic starting point. And in fact, to some extent, I agree with some of your statements about NR because a lot of current sensors do at the individual CMOS sensel level something that resembles what used to be a full CCD calibration (I base this opinion on the relatively detailed - but nowhere near what Kodak provides for its scientific sensors - information Sony provides on its sensors - links previously given on this board).

But, assuming your statements contain a tiny bit of truth, it would be nice if you could develop your argument without resorting to considerations about atmospheric composition. In this very thread, ondebanks and I slightly disagreed - have you noticed that neither of us ended up attempting to ridicule the other and that both of us presented verifiable arguments? Is that something you would consider? Disagreement, and particularly disagreement about the spoken or written word, is typical of us humans. Learn to live with it and accept, when the object of the discussion is amenable to experimental investigations, that demonstrations/experimentation are better than aggressive discourse.

Last but not least, try to be modest in your claims. About 15 years ago, I build a CCD camera and programmed its firmware from scratch (and litterature, of course, I did not invent the thing). At that time, I certainly felt that I knew a lot about how sensors worked. Today is a very different era in sensing techonology and, except for people who devote their whole life to a specific sub-area of this immense topic, no one can claim to know everything that goes on under the bonnet.

AFAIC, even if I don't always agree 100% with a competent person (such as Emil, Ondebanks, Bart and many others), I am grateful for what they share and I see the eventual friction points between what they say or write and what I think as opportunities to improve my understanding.
Title: Re: CCD and CMOS
Post by: madmanchan on September 30, 2011, 07:54:10 AM
I think the argument is based on the camera testing at the end of manufacture. It is going to be tested against known images of blue , green, red, for output accuracy. The firmware is set to correct any deviation in the signal. In CMOS the correction is against each pixel. In CCD the conversion of photon to current is against whole rows so the converter is calibrated against 1000s of pixels output. Of course each pixel still has to output the same value so the correction of the color is the same. The accuracy of the signal converter should be much greater in CCD.

I think these are minor differences in the big picture. 

By far the biggest factor in the output color response is the spectral transmission of the Bayer color filters.  As far as color reproduction goes (*), I've measured some CCDs with great spectral transmission curves and other CCDs with not-so-great ones.  The same with CMOS sensors. 

Eric

(*) i.e., getting a good match on the so-called Luther-Ives condition
Title: Re: CCD and CMOS
Post by: Daniel Browning on September 30, 2011, 11:26:47 AM
Emil and Jerry may be very knowledgeable and careful workers, but a FFT analysis of a 30 second Darkframe from the D3 is shown below and I see no evidence of filtering.

The only purpose that Nikon intended for the so-called noise reduction is to suppress unmapped hot/dead pixels, so the effect is definitely going to be a lot more subtle than a global median filter. That's why I prefer the astrophotographer's term for it, the "Nikon star killer algorithm." :)

A highly talented engineer named Marianne Oelund completed a rather detailed characterization of the algorithm as it was implemented in older Nikon cameras, the D3, and most recently, the D7000:

http://forums.dpreview.com/forums/read.asp?forum=1021&message=34309201

Here is a link to an image demonstrating the star killer effect:

http://actionphotosbymarianne.com/spectra/HPSeffectEx1.jpg
Title: Re: CCD and CMOS
Post by: bjanes on September 30, 2011, 12:40:57 PM
The only purpose that Nikon intended for the so-called noise reduction is to suppress unmapped hot/dead pixels, so the effect is definitely going to be a lot more subtle than a global median filter. That's why I prefer the astrophotographer's term for it, the "Nikon star killer algorithm." :)

A highly talented engineer named Marianne Oelund completed a rather detailed characterization of the algorithm as it was implemented in older Nikon cameras, the D3, and most recently, the D7000:

http://forums.dpreview.com/forums/read.asp?forum=1021&message=34309201

Here is a link to an image demonstrating the star killer effect:

http://actionphotosbymarianne.com/spectra/HPSeffectEx1.jpg

Daniel,

Thanks for the links. I did look at them and found them quite interesting. Marianne is truly amazing--it is unfortunate that she does not post on LuLa. Emil did contribute to that thread, and he noted that the effect did not show up on Fourier analysis, so my tests likely were not sufficiently sensitive to detect hot pixel filtering.

Regards,

Bill
Title: Re: CCD and CMOS
Post by: Fine_Art on September 30, 2011, 04:51:45 PM
I think these are minor differences in the big picture. 

By far the biggest factor in the output color response is the spectral transmission of the Bayer color filters.  As far as color reproduction goes (*), I've measured some CCDs with great spectral transmission curves and other CCDs with not-so-great ones.  The same with CMOS sensors. 

Eric

(*) i.e., getting a good match on the so-called Luther-Ives condition

They are small in the sense that all consumer cameras now have very good color reproduction, CCD or CMOS. Its still a consumer product not something made for NASA or a defense department. The tolerance in the parts is a real issue when there is a very competitive market. For example Pentax in the K5 managed to squeeze much more out of the same sensor than Sony, the manufacturer, or Nikon did with their version of the camera. They chose some higher quality part. With CCD row conversion a higher quality signal converter has much more return per part due to it's effect on entire rows.

Maybe you are right that the tolerances of the color filters have much more effect. Anyone know how they degrade over time?
Title: Re: CCD and CMOS
Post by: ErikKaffehr on September 30, 2011, 05:08:05 PM
Hi,

I would leave out NASA, they do a lot of false color work and also a lot of invisible part of electromagnetic spectrum stuff. Also NASA stuff is probably radiation hardened.

The main reason that Sony is lagging Pentax and Nikon may be that their ASIC (Bionz) is 12 bits, while Pentax and Nikon use 14 bits.

If you compare the figures you see that all cameras are essentially on the same line regarding SNR. Pentax has slightly better SNR at high ISO due to raw noise reduction.

DR is worse on the sony, that could be read noise or 12-bitness.

Best regards
Erik




They are small in the sense that all consumer cameras now have very good color reproduction, CCD or CMOS. Its still a consumer product not something made for NASA or a defense department. The tolerance in the parts is a real issue when there is a very competitive market. For example Pentax in the K5 managed to squeeze much more out of the same sensor than Sony, the manufacturer, or Nikon did with their version of the camera. They chose some higher quality part. With CCD row conversion a higher quality signal converter has much more return per part due to it's effect on entire rows.

Maybe you are right that the tolerances of the color filters have much more effect. Anyone know how they degrade over time?
Title: Re: CCD and CMOS
Post by: ondebanks on September 30, 2011, 08:00:20 PM
I would leave out NASA, they do a lot of false color work and also a lot of invisible part of electromagnetic spectrum stuff. Also NASA stuff is probably radiation hardened.

Absolutely. I would be surprised if NASA ever used a stills imager with a CFA, other than "reportage" cameras to document launches and stuff happening on manned missions. Imagers for their astronomy & space science programmes would not want to have a fixed, predetermined colour filtration, which cuts out valuable UV and deep red/IR wavelengths, and where maybe only 1/4 of the pixels are sensitive to what's being measured. I can't speak for every mission, but I've worked a lot with Hubble data and I can guarantee that none of the UV (FOC, STIS), Visible (WF/PC, WFPC2, ACS) or IR (NICMOS) cameras had a CFA.

Ray
Title: Re: CCD and CMOS
Post by: ondebanks on September 30, 2011, 08:20:32 PM
I think these are minor differences in the big picture. 

By far the biggest factor in the output color response is the spectral transmission of the Bayer color filters.  As far as color reproduction goes (*), I've measured some CCDs with great spectral transmission curves and other CCDs with not-so-great ones.  The same with CMOS sensors. 

Eric

(*) i.e., getting a good match on the so-called Luther-Ives condition

Eric,

Apropos of nothing, you might be amused to hear that for some time, whenever I saw your username "madmanchan" I kept reading it as "Mad Manchán" - Manchán is an old and uncommon Irish forename (pronounced Man-KHAWN). Your signature alerted me to the fact that it's really "Madman Chan"!  :D

Anyway, I'm interested in your statement that you've measured several sensor transmission curves. What's your measurement setup? Have you seen any significant deviations from the corresponding curves in data sheets?

Ray

Title: Re: CCD and CMOS
Post by: hjulenissen on October 01, 2011, 02:23:20 PM
How much "better" would/could the color response have been if one removed the CFA and used a set of 3 purpose-made color filters that were inserted one at a time for 3 separate exposures (assuming that the scene did not run and hide in-between exposures)?

So perhaps the question could be paraphrased "how big a limitation is it that the spectral filtering carried out in the CFA have to have really small features, and be economically/practically feasible"?

I believe that color wheels are commonly used for multi-spectral cameras. I would guess that with 7 or 10 wisely chose bandpass filters, you would have a lot of options not available to regular cameras.

-h
Title: Re: CCD and CMOS
Post by: PierreVandevenne on October 01, 2011, 02:44:25 PM
Well, most medium to upper level amateur astro cameras come with filter wheels. You typically take three exposures, and then combine them in a color image. Here's a typical filter wheel
http://www.optcorp.com/product.aspx?pid=319-327-339-4242
Here's a typical integrated wheel
http://www.qsimaging.com/540-overview.html

The differences with a RGB filter array is, of course, that when one filter is in front of the camera, you benefit from photons captured by all the sensels, thereby doubling the green practical QE vs a bayer matrix and quadrupling red and blue. Thise comes, of course, with the penalty of having to do three exposures. The problems for photography are essentially that conditions change when you take frames in succession: the camera moves a very tiny bit, the lighting has changed, etc... and also that you end up with three images that aren't identical. Even the focal plane can change its position somewhat (it's dramatic in an achromat, tolerable in an apochromat). Assuming a star is red, you'll also get, even if the focus is perfect, a larger diameter in the red channel that you will in the green channel and you'll have to handle that in some way. It's very bad for bright point sources, maybe a bit less for full images with less luminosity differences in terms of photography, but you don't want to have to deal with all those issues shooting pictures.
The way those cameras are used, when they aren't used with standard RGB filters which have no other purpose than producing "pretty pictures" of no scientific value, is with filters whose bandwidth is very well defined (http://en.wikipedia.org/wiki/Photometric_system)
BTW, that tri-color filter processed was used in the early 20th century. http://en.wikipedia.org/wiki/Sergei_Mikhailovich_Prokudin-Gorskii

And of course, a variation of that is the triple CCD in some video cameras
Title: Re: CCD and CMOS
Post by: theguywitha645d on October 01, 2011, 02:46:08 PM
I run a variety of microscope cameras with color filters (commonly the same filtration as the Bayer filters) and Bayer patterns. The color is practically speaking the same. The only advantage to the filtered camera is the possibility of tuning the filters--color fitter wheels are a little old fashioned and LCD tunable filters work better. However, with broad bands, there is not much benefit in tuning the filters. That is usually left to very narrow bands measured in angstrom.

As far as the loss of resolution due to the Bater pattern, it is insignificant. The benefit of unfiltered monochrome sensors is really in sensitivity.
Title: Re: CCD and CMOS
Post by: madmanchan on October 01, 2011, 04:54:30 PM
As far as the sensor is concerned, the important factor (as noted above by others) in color reproduction is not the density of the color filters (or their spatial arrangement in a mosaic pattern such as Bayer), but rather the shapes of the transmission curves and how they relate to each other. Ideally from a color perspective, you'd want the transmission curves to be the same as the human cone responses (in the eye) or a linear transformation thereof. But there is a tradeoff in terms of color vs noise, and of course there are other practical constraints due to materials, manufacturing, costs, etc., so in practice this technical condition is not satisfied. as I mentioned earlier, this is rather a separate issue from the choice of CCD vs CMOS.

But as far as photography is concerned, my experience has been that color response differences from system to system have less to do with the sensor, and more to do with the software rendering applied in post-processing (even if the user never touches any sliders or controls). Example: Canon has various Picture Styles (such as Portrait and Landscape) available in their software for their cameras, some of which have CMOS sensors, some of which have CCD sensors. The difference in visual appearance between these software-based styles is far greater than the actual differences in the color filters!

Ray, I generally measure camera optical systems with a monochromator to estimate the transmission curves over the visible and near-IR range. However, I don't have manufacturer data sheets for most of the systems I measure (and even for those for which I do, the maker's data is usually for the sensor alone, whereas I prefer to measure sensor + lens combinations, so comparisons are hard). And you never know -- maybe I was a mad Irishman in a previous life!! ;D
Title: Re: CCD and CMOS
Post by: hjulenissen on October 03, 2011, 03:08:13 AM
And of course, a variation of that is the triple CCD in some video cameras
But then each photon is counted (at least in theory). For bayer and color wheel solutions, only e.g. 1/3rd of the photons hitting the sensor during the total exposure time is counted, the rest is absorbed in spectral bandpass filters.

-h
Title: Re: CCD and CMOS
Post by: hjulenissen on October 03, 2011, 03:10:49 AM
As far as the sensor is concerned, the important factor (as noted above by others) in color reproduction is not the density of the color filters (or their spatial arrangement in a mosaic pattern such as Bayer), but rather the shapes of the transmission curves and how they relate to each other.
Sure, but my gut-feeling is that whenever you have to do something really tiny, complex and economical, you loose something. If that gut-feeling is wrong, and Canon & Nikon are free to make whatever spectral response they see fit (keeping in mind the color response vs noise issue you mentioned), then my gut-feeling was wrong.

-h
Title: Re: CCD and CMOS
Post by: PierreVandevenne on October 03, 2011, 07:36:06 AM
But then each photon is counted (at least in theory). For bayer and color wheel solutions, only e.g. 1/3rd of the photons hitting the sensor during the total exposure time is counted, the rest is absorbed in spectral bandpass filters.

Not sure about that

http://en.wikipedia.org/wiki/File:Dichroic-prism.svg

http://en.wikipedia.org/wiki/File:A_3CCD_imaging_block.jpg

I'd don't have a well defined opinion on the efficiency of splitting vs filtering, but I am under the impression the sensors in a 3CCD or 3MOS cameras don't get all the photons. If they did, it would be a mess to colour balance imho.
Title: Re: CCD and CMOS
Post by: eronald on October 03, 2011, 08:25:25 AM
Not sure about that

http://en.wikipedia.org/wiki/File:Dichroic-prism.svg

http://en.wikipedia.org/wiki/File:A_3CCD_imaging_block.jpg

I'd don't have a well defined opinion on the efficiency of splitting vs filtering, but I am under the impression the sensors in a 3CCD or 3MOS cameras don't get all the photons. If they did, it would be a mess to colour balance imho.

+1

Edmund
Title: Re: CCD and CMOS
Post by: hjulenissen on October 03, 2011, 08:45:05 AM
Not sure about that

http://en.wikipedia.org/wiki/File:Dichroic-prism.svg

http://en.wikipedia.org/wiki/File:A_3CCD_imaging_block.jpg

I'd don't have a well defined opinion on the efficiency of splitting vs filtering, but I am under the impression the sensors in a 3CCD or 3MOS cameras don't get all the photons. If they did, it would be a mess to colour balance imho.
How should I interpret those figures in light of you statement?

I am by no means an expert on this topic. But it seems to me that if such a thing as "perfect" splitting of light based on wavelength exists (I am sure that it does not, but perhaps as an approximation), then some kind of 3-band bandpass filtering might be possible with a "3CCD" solution. It might not provide the _desirable_ shape of spectral selectivity, and it may have all kinds of practical/economical drawbacks, but I think this is an interesting aspect of it.

It all boils down to doing spectral selection using spectral absorption vs spectral reflectance - at least on the level of physics that I am able to follow :-)

-h
Title: Re: CCD and CMOS
Post by: PierreVandevenne on October 03, 2011, 10:27:14 AM
It seems that multi channel dichroic prisms are, in theory at least, better than wideband RGB filters, in the sense that there are no holes, spikes, overlaps, etc... in the transmission band. The incoming light is split, you characterize it and that's it. I guess this could also allow for different distances for the three focal planes to compensate for chromatic aberration.

http://www.optec.eu/eng/multichannel/1194.htm

http://www.firstlightoptics.com/rgb-filters-filter-sets/baader-lrgbc-ccd-filter-set.html

But in practice, I have only worked with wide and narrow band filters and therefore will try to keep my foot out of my mouth, waiting for someone more competent in those matters to eventually jump in ;-)
Title: Re: CCD and CMOS
Post by: bjanes on October 03, 2011, 10:31:05 AM
As far as the sensor is concerned, the important factor (as noted above by others) in color reproduction is not the density of the color filters (or their spatial arrangement in a mosaic pattern such as Bayer), but rather the shapes of the transmission curves and how they relate to each other. Ideally from a color perspective, you'd want the transmission curves to be the same as the human cone responses (in the eye) or a linear transformation thereof. But there is a tradeoff in terms of color vs noise, and of course there are other practical constraints due to materials, manufacturing, costs, etc., so in practice this technical condition is not satisfied. as I mentioned earlier, this is rather a separate issue from the choice of CCD vs CMOS.

An example of these tradeoffs is discussed in the DXO paper (http://www.dxomark.com/index.php/Publications/DxOMark-Insights/Canon-500D-T1i-vs.-Nikon-D5000/Color-blindness-sensor-quality) comparing the Nikon D5000 with the Canon EOS 500D. The Canon has poor color depth due the characteristics of its CFA filters. The problem lies mainly in the Red CFA filter, which is actually more sensitive to green rather than red as shown below. This necessitates a large coefficient in the color matrix, which adds noise. In contrast, the Nikon has a better red response and a greater color depth.

CCD sensors can also have unfavorable CFA characteristics as shown by the DXO analysis of the Phase One P45+, where the red channel is also more sensitive to green than red. The camera has a poor metamerism index 0f 72, as compared to an index of 83 for the D5000. The P45+ is an older camera, and the situation is much improved with the newer P40+. These studies indicate that CCDs do not necessarily have better color depth than CMOS designs.

Regards,

Bill

Title: Re: CCD and CMOS
Post by: hjulenissen on October 03, 2011, 11:05:57 AM
An example of these tradeoffs is discussed in the DXO paper (http://www.dxomark.com/index.php/Publications/DxOMark-Insights/Canon-500D-T1i-vs.-Nikon-D5000/Color-blindness-sensor-quality) comparing the Nikon D5000 with the Canon EOS 500D. The Canon has poor color depth due the characteristics of its CFA filters. The problem lies mainly in the Red CFA filter, which is actually more sensitive to green rather than red as shown below. This necessitates a large coefficient in the color matrix, which adds noise. In contrast, the Nikon has a better red response and a greater color depth.
Do you think that this is a trade-off of achromatic SNR vs color noise, or sensor cost/performance vs color noise?

-h
Title: Re: CCD and CMOS
Post by: bjanes on October 03, 2011, 11:57:26 AM
Do you think that this is a trade-off of achromatic SNR vs color noise, or sensor cost/performance vs color noise?

-h

The article states, "This comparison is a bit surprising with respect to the previous SNR 18% results. Why such a difference? Color sensitivity is impacted by noise curves and spectral responses. If SNR curves are close, most of the divergence observed must be due to a difference in spectral sensitivities, which implies very different color processing for each sensor."

I conclude that the difference is largely due to color noise.

Regards,

Bill
Title: Re: CCD and CMOS
Post by: hjulenissen on October 03, 2011, 12:41:57 PM
The article states, "This comparison is a bit surprising with respect to the previous SNR 18% results. Why such a difference? Color sensitivity is impacted by noise curves and spectral responses. If SNR curves are close, most of the divergence observed must be due to a difference in spectral sensitivities, which implies very different color processing for each sensor."

I conclude that the difference is largely due to color noise.

Regards,

Bill
I should have phrased my question differently. Given that Canon have less spectrally selective CFA than Nikon, and thereby a color correction matrix that is more different from the identity matrix and more color-noise prone:
-Did they do this because they think that having wider filters, passing more photons, gives them an advantage when shooting spectrally broad/flat scenes
-Or does Canon have a sensor with a disadvantage in the first place, and spectrally wide filters used to hide its flaws

Or perhaps this is a feature to the silicon process that Canon use, linked perhaps to micro lenses etc?

I have heard that Sony alpha DSLRs have a radically different philosophy (closer to the standard CIE observer, at the cost of more noise)?

-h
Title: Re: CCD and CMOS
Post by: ErikKaffehr on October 03, 2011, 01:17:10 PM
Hi,

I guess that Canon seeks better high ISO performance and therefore has more overlap between the filters in the CGA. But, that is just a guess.

Best regards
Erik

I should have phrased my question differently. Given that Canon have less spectrally selective CFA than Nikon, and thereby a color correction matrix that is more different from the identity matrix and more color-noise prone:
-Did they do this because they think that having wider filters, passing more photons, gives them an advantage when shooting spectrally broad/flat scenes
-Or does Canon have a sensor with a disadvantage in the first place, and spectrally wide filters used to hide its flaws

Or perhaps this is a feature to the silicon process that Canon use, linked perhaps to micro lenses etc?

I have heard that Sony alpha DSLRs have a radically different philosophy (closer to the standard CIE observer, at the cost of more noise)?

-h
Title: Re: CCD and CMOS
Post by: mikejyg on January 11, 2013, 12:10:50 PM
I think there is still one important area that CMOS can not approach CCD's quality: uniformity.

Its effect is subtle and not easily understood, and CMOS manufacturers do not like to talk about it either.

It's like the THD (dynamic noise) in audio quality, vs SNR (the common term "noise" people use in describing CMOS and CCD).
Title: Re: CCD and CMOS
Post by: Fine_Art on January 11, 2013, 05:07:06 PM
Hi,

I guess that Canon seeks better high ISO performance and therefore has more overlap between the filters in the CGA. But, that is just a guess.

Best regards
Erik


That was my take as well. It's for high ISO performance by letting through more photons.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 11, 2013, 05:17:30 PM
It seems that multi channel dichroic prisms are, in theory at least, better than wideband RGB filters, in the sense that there are no holes, spikes, overlaps, etc... in the transmission band. The incoming light is split, you characterize it and that's it. I guess this could also allow for different distances for the three focal planes to compensate for chromatic aberration.

http://www.optec.eu/eng/multichannel/1194.htm

http://www.firstlightoptics.com/rgb-filters-filter-sets/baader-lrgbc-ccd-filter-set.html

But in practice, I have only worked with wide and narrow band filters and therefore will try to keep my foot out of my mouth, waiting for someone more competent in those matters to eventually jump in ;-)


You know a lot more about the technology than me.

What I will say is I have seen 3 chip HD video vs 1 chip HD video. The 3 chip systems look way better. Maybe an order of magnitude better. Go to your local electronics store and compare the cameras for yourself. Panasonic makes a nice 3CMOS camcorder. Compare it to any manufacturer using 1 chip of similar size. Not the sony nex, that is a much bigger chip.

Edit: by compare I mean shoot video in the store with each. Output it to a HDTV.
Title: Re: CCD and CMOS
Post by: jeremypayne on January 11, 2013, 05:21:30 PM
I think there is still one important area that CMOS can not approach CCD's quality: uniformity.

Hi ... can you back that up? 

Never heard that before and not sure what would account for such a difference.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 11, 2013, 06:20:52 PM
Hi ... can you back that up? 

Never heard that before and not sure what would account for such a difference.

Its old tech. It used to be that the CCDs were making better color uniformity. CMOS looked artificial.

Many people thought it was based on the method of charge conversion. CMOS integrated the A/D converter into the pixel. CCD sent all the charge to a row converter. The theory was the CCD method allowed calibration of the value to a standard with higher tolerance than a single value from a pixel.

The last generation of CMOS from all the DSLR manufacturers seems to have much better color accuracy than a few years ago. Pictures now look realistic, like looking through a window whereas the CMOS a few years back looks a bit artificial. Maybe its improved bayer algorithms. Maybe its the latest updates to RAW converters. Something has changed on the CMOS side.
Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 12, 2013, 01:08:48 AM
Hi,

CMOS never integrated the ADC into a pixel. The major difference is that CMOS essentially outputs a voltage while CCDs shift the charge of each pixel to the edge of the sensor. CMOS needs a few gates to handle readout in each pixel while CCD has no electronics at all.

On older Nikons the ADC-s were off chip, they were using Burr&Brown converters in external housings. I'm pretty sure that Canon also uses off chip converters.

The most significant difference between CCD and CMOS may be that CMOS can be read out nondestructively. So, essentially all CMOS vendor use a method called correlated double sampling where they measure charge on each pixel after reset and by that method are able to eliminate much of the noise.

Best regards
Erik

Its old tech. It used to be that the CCDs were making better color uniformity. CMOS looked artificial.

Many people thought it was based on the method of charge conversion. CMOS integrated the A/D converter into the pixel. CCD sent all the charge to a row converter. The theory was the CCD method allowed calibration of the value to a standard with higher tolerance than a single value from a pixel.

The last generation of CMOS from all the DSLR manufacturers seems to have much better color accuracy than a few years ago. Pictures now look realistic, like looking through a window whereas the CMOS a few years back looks a bit artificial. Maybe its improved bayer algorithms. Maybe its the latest updates to RAW converters. Something has changed on the CMOS side.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 12, 2013, 02:01:42 AM
We both use Sony and they have advertised on chip A/D conversion for ages.

Yes, Canon and Nikon were doing something different, maybe why they had 14 bit while we were stuck with 12. Pentax also tried high end external 22 bit converters. They went back to something basic in the next version.
Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 12, 2013, 03:16:00 AM
Hi,

To be exact, Nikon D800/D800E, D600 and D3X use Sony Exmoor sensors using on chip converters. D3/D3S and D4 uses Nikon designed chips using external converters. Nikon D3X had both 14 bit readout (slow 2FPS) and 12 bit readout. I guess the reason Sony had 12 bit converters was more related to Bionz being a 12 bit chip. The Alpha 77 I have now has 14 bit conversion.

Pentax used Samsung chips in older cameras the K5 has an Exmoor with 14 bit readout.

DR figures from DxO (latest sensor). The DxO figures are scaled for a smallish print size. DR for actual pixels cannot exceed the channel width.

CameraDRcomments
Nikon D80014.4Exmoor based sensors
Nikon D60014.2
Sony RX114.3
PentaxK514.1Note that K5 is APS-C, impressive feat!
Sony Alpha 99 (SLT) 14
Phase One IQ18013.6Best medium format CCD based camera
Nikon D413.1Best non Exmoor design CMOS
Canon EOS 1Dx11.8
Leica M911.7Best CCD based full frame

Updated: Small corrections + added Phase One IQ180 replacing Pentax 645D as best CCD MF camera.
Best regards
Erik


We both use Sony and they have advertised on chip A/D conversion for ages.

Yes, Canon and Nikon were doing something different, maybe why they had 14 bit while we were stuck with 12. Pentax also tried high end external 22 bit converters. They went back to something basic in the next version.
Title: Re: CCD and CMOS
Post by: FranciscoDisilvestro on January 12, 2013, 03:30:01 AM

DR figures from DxO (latest sensor)

CameraDRExmoor based sensors
Nikon D80014.4
Nikon D60014.2
Sony RX114.3
PentaxK514.1
Sony Alpha 99 (SLT) 14
Nikon D413.1Best non Exmoor design
Pentax 645D12.6Best CCD based camera
Canon EOS 1Dx11.8
Leica M911.7Best CCD based full frame

Best regards
Erik



Hi, Remember those figures are for the print performance (imaged resized to 8"x12" @ 300 dpi)
It is not possible to have more than 14 bit DR at the sensor with 14 bit linear encoding

The highest DR (according to DxOMark) would be thge Sony RX1 with 13.47 and Nikon D600 with 13.44

regards,
Francisco
Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 12, 2013, 03:50:35 AM
Hi,

I remember that those values are scaled for same size but I actually feel that they are more relevant than actual pixel values, also they were much easier to find.

You are absolutely right that it is not possible to encode more than 14 EV in 14 bit linear space.

The interesting point is that all the top cameras use Sony technology. Nikon uses Toshiba for D5200 by the way. I guess Sony is not alone with on chip per column conversion. The new sensor for the Leica M is also a CMOS design with column type on chip converters.

http://www.cmosis.com/news/press_releases/new_leica_m_uses_cmosis_24_mp_cmos_image_sensor

http://www.chipworks.com/blog/technologyblog/2012/10/25/full-frame-dslr-cameras-part-iii-new-entrants-and-look-forward/

It will be interesting to compare Sony's technology with similar constructions from the competition.

Best regards
Erik


Hi, Remember those figures are for the print performance (imaged resized to 8"x12" @ 300 dpi)
It is not possible to have more than 14 bit DR at the sensor with 14 bit linear encoding

The highest DR (according to DxOMark) would be thge Sony RX1 with 13.47 and Nikon D600 with 13.44

regards,
Francisco
Title: Re: CCD and CMOS
Post by: jeremypayne on January 12, 2013, 09:21:47 AM
It used to be that the CCDs were making better color uniformity. CMOS looked artificial.

Previously, you stated that better "uniformity" was "still" an advantage for CCD-based sensors.

I don't think there is anything inherent in CCD technology that would account for that.
Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 12, 2013, 11:13:37 AM
Hi,

I would suggest that there are a lot of myths in this area. CMOS is main stream. CCDs are used by small companies making high end stuff. Most CCDs are made by Kodak (R.I.P.) or Dalsa.  In general, CCDs often lack OLP filtering which is good for perceived sharpness but creates a lot of fake detail.

Kodak and Dalsa may use a different CGA (Color Grid Array) than say Nikon or Canon. So the myth goes that CCD has better color then CMOS, but that has little to do with CCD vs CMOS and a lot to do with CGA design.

Best regards
Erik




Previously, you stated that better "uniformity" was "still" an advantage for CCD-based sensors.

I don't think there is anything inherent in CCD technology that would account for that.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 12, 2013, 03:17:11 PM
Previously, you stated that better "uniformity" was "still" an advantage for CCD-based sensors.

I don't think there is anything inherent in CCD technology that would account for that.

I believed it until I started seeing a lot of pictures from people using the Sony A99, A77 ( I use sony), the Nikon D800, D600.
Maybe its the down-sampling leeway of the 24MP of data. Maybe its improved software. I dont know. All I know is these cameras are outputting images that look very realistic vs older generation DSLRs.

The DR and the color tone accuracy are improved, from hardware or software is unknown.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 12, 2013, 03:21:30 PM
Its not just my eyes, DxO is measuring a jump in image quality on these cameras.
Title: CCD and CMOS: Panasonic also doing on chip per column conversion
Post by: BJL on January 12, 2013, 04:05:02 PM
I guess Sony is not alone with on chip per column conversion. The new sensor for the Leica M is also a CMOS design with column type on chip converters.

http://www.cmosis.com/news/press_releases/new_leica_m_uses_cmosis_24_mp_cmos_image_sensor

http://www.chipworks.com/blog/technologyblog/2012/10/25/full-frame-dslr-cameras-part-iii-new-entrants-and-look-forward/
Panasonic is also doing column-parallel ADC in some sensors: read about the MN34070 is this document
http://www.semicon.panasonic.co.jp/en/support/catalogs/pdf/T12013CE.pdf

Chipworks has confirmed that this is the sensor in the GH2:
https://chipworks.secure.force.com/catalog/ProductDetails?sku=PAN-DMC-GH2-K_Pri-Camera
Title: Re: CCD and CMOS
Post by: jeremypayne on January 12, 2013, 05:16:42 PM
The DR and the color tone accuracy are improved, from hardware or software is unknown.

No doubt... But not the question at hand.

The point is ... there are not inherent characterisics of CCD technology that produce "more uniform" output relative to CMOS technology.

Any differences are due to other factors and design choices not specific to CCD vs CMOS.

As alluded to by Erik  ... This would qualify as a bit of an urban legend.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 12, 2013, 05:24:36 PM
Its more a case of CCD tech was far more mature than CMOS tech. Now CMOS is as good or better on all parameters.
Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 13, 2013, 02:09:20 AM
Hi,

Much of the image quality actually comes from the raw converters. For instance the new processing pipe line introduced in LR 3 was a major improvement at least for my Alpha 99, see here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/32-new-raw-processing-pipeline-in-lightroom-3-beta-2

I have recently acquired a Sony Alpha 99. Naturally enough I shot some comparison pictures with my Alpha 900 and could see no difference at low ISO. I did shoot a horse jumping at ISO 6400 and printed A3, something I wouldn't dream of with the Alpha 900, so I'm pretty sure the Alpha 99 has a cleaner sensor (less readout noise and probably a higher FWC).

The main reason I bought the Alpha 99 is live view.

Best regards
Erik


I believed it until I started seeing a lot of pictures from people using the Sony A99, A77 ( I use sony), the Nikon D800, D600.
Maybe its the down-sampling leeway of the 24MP of data. Maybe its improved software. I dont know. All I know is these cameras are outputting images that look very realistic vs older generation DSLRs.

The DR and the color tone accuracy are improved, from hardware or software is unknown.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 13, 2013, 12:08:31 PM
It does look like a big improvement in LR3 over past versions. I had never been impressed with the product before, maybe now it's worth a look. I still see chroma smear over the image.

The boost in fine detail contrast looks like a fine wavelet pass.
Title: Re: CCD and CMOS
Post by: jeremypayne on January 13, 2013, 01:45:29 PM
It does look like a big improvement in LR3 over past versions.

Lightroom is on v4.x

Lightroom is beyond impressive to me ... it the single most impressive pieces of "consumer" software I have ever seen.  The ambition and "game-changing" nature of the project is inspiring to me as someone who lives in "legacy" world trying to innovate.

The story of Lightroom is a an amazing story.
Title: Re: CCD and CMOS
Post by: Fine_Art on January 13, 2013, 02:23:16 PM
Lightroom is on v4.x

Lightroom is beyond impressive to me ... it the single most impressive pieces of "consumer" software I have ever seen.  The ambition and "game-changing" nature of the project is inspiring to me as someone who lives in "legacy" world trying to innovate.

The story of Lightroom is a an amazing story.

Lets not get carried away. We are not talking about a Saint here.
Title: Re: CCD and CMOS
Post by: jeremypayne on January 13, 2013, 03:14:00 PM
Lets not get carried away. We are not talking about a Saint here.

I stand by every word I said. 

I manage a lot of software development as part of my professional life.  One of the "catchphrases" I use often in the context of user-driven software companies and projects ... "Rather than build a little something for everyone, build everything for someone."

The Lightroom team clearly understands that principle.
Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 13, 2013, 03:22:41 PM
Hi,

For me Lightroom was exactly the program I wanted to write, but I estimated it at 30-100 man years, longer than I plan to live.

Best regards
Erik


I stand by every word I said. 

I manage a lot of software development as part of my professional life.  One of the "catchphrases" I use often in the context of user-driven software companies and projects ... "Rather than build a little something for everyone, build everything for someone."

The Lightroom team clearly understands that principle.
Title: Re: CCD and CMOS
Post by: eronald on January 14, 2013, 10:18:47 PM
From my perspective, Lightroom is a very nice light-table, filing and printing program with a good-enough Raw converter. If you want superlative Raw conversion, and only raw conversion, you can often get it for free with the manufacturer's software eg. Canon DPP, freeware like RPP, or resort to boutique products like Iridient's Raw Developer, or even Capture One which I believe quite a few people on this forum have used.

Edmund

Hi,

For me Lightroom was exactly the program I wanted to write, but I estimated it at 30-100 man years, longer than I plan to live.

Best regards
Erik


Title: Re: CCD and CMOS
Post by: ErikKaffehr on January 15, 2013, 12:43:51 AM
Hi,

I was developing a slide presentation program called slide show at that time. It was essentially a database driven program for handling and keywording large amounts of images. The program could generate slide shows, it would put images in a folder together with an XML file telling the program what to do with them. I added some options to the XML to adjust small issues found in replay. I than realized I could add raw conversion and do most in processing in my program. But that would involve a lot, really learning about color spaces and management and so. I also guess it's easy to mess up.

Than Lightroom came along, and did things in mostly the way I wanted.

The way it is my preference is pretty much to work parametrically. I don't want to go over 16-bit TIFFS. I also much prefer to do most of the work in Lightroom, I don't raw conversion and process in Photoshop, unless needed.

Since "Madman Chen" is on the ACR team, things go in the right direction. Now we have lens corrections, can do keystone corrections in develop, and we also have some tone mapping capabilities (although the latter are poorly described and documented).

Best regards
Erik





From my perspective, Lightroom is a very nice light-table, filing and printing program with a good-enough Raw converter. If you want superlative Raw conversion, and only raw conversion, you can often get it for free with the manufacturer's software eg. Canon DPP, freeware like RPP, or resort to boutique products like Iridient's Raw Developer, or even Capture One which I believe quite a few people on this forum have used.

Edmund