Pages: [1] 2   Go Down

Author Topic: Canon 5D III on chip noise reduction?  (Read 8897 times)

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Canon 5D III on chip noise reduction?
« on: March 02, 2012, 07:59:26 pm »

Does any one have any thoughts on DP reviews statement:
 "Canon has also applied its gapless microlens array, meaning more of the light hitting the sensor gets directed down into the photodiodes. Finally there's a newly-developed on-chip noise reduction system to improve the quality of the information coming off the chip. Canon isn't making specific claims for how much of an improvement these changes make to the raw output but, once subjected to the Digic 5+'s processing, it will claim a 2-stop improvement in the JPEG images"
Any improvement in DR?
Marc
Logged
Marc McCalmont

Kirk Gittings

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1561
    • http://www.KirkGittings.com
Re: Canon 5D III on chip noise reduction?
« Reply #1 on: March 02, 2012, 08:10:18 pm »

I have that exact same question.
Logged
Thanks,
Kirk Gittings

bill t.

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3011
    • http://www.unit16.net
Re: Canon 5D III on chip noise reduction?
« Reply #2 on: March 02, 2012, 08:57:10 pm »

Well as someone else pointed out on another thread, noise and dynamic range are inversely related.  The less noise there is, the more information you can fish out of the dark parts of the raw data.  So if the 5D3 has less noise than some other DSLR, it probably has more dynamic range.  I was very impressed with that effect going from a D2X to the 5D2, it was almost like single 5D2 shots were the dynamic range equivalent of a +-1 stop HDR set on the D2X.  So if that same ratio of improvement applies for 5D2->5D3, I might just pick one up!

Also I would like to know what all this "multiple exposure" voodoo is that is ambiguously mentioned in the 5D3 hype. Does that perhaps mean exposure stacking, bracketing, HDR, etc is somehow made automatic, and hopefully very fast?  For us tripod shooters that seems to hold out the possibility of easily averaging several exposures, further pushing down noise and increasing useable image data.
Logged

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Re: Canon 5D III on chip noise reduction?
« Reply #3 on: March 02, 2012, 09:08:36 pm »

Also I would like to know what all this "multiple exposure" voodoo is that is ambiguously mentioned in the 5D3 hype. Does that perhaps mean exposure stacking, bracketing, HDR, etc is somehow made automatic, and hopefully very fast?  For us tripod shooters that seems to hold out the possibility of easily averaging several exposures, further pushing down noise and increasing useable image data.

I interpret it as in camera HDR but is it just jpeg or RAW HDR?
Marc
Logged
Marc McCalmont

Kirk Gittings

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1561
    • http://www.KirkGittings.com
Re: Canon 5D III on chip noise reduction?
« Reply #4 on: March 02, 2012, 09:12:11 pm »

I find it interesting that I have seen no examples of the HDR capability posted. I would think they would be pushing that pretty hard.
Logged
Thanks,
Kirk Gittings

mmurph

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 506
    • http://
Re: Canon 5D III on chip noise reduction?
« Reply #5 on: March 02, 2012, 10:01:04 pm »

The multiple exposure images use a variation of the image stacking process.

I know one commentator opened up the shadows quite a bit and commented on the low noise level. So hopefully they are using the multiple images to subtract out noise?

Michael
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: Canon 5D III on chip noise reduction?
« Reply #6 on: March 03, 2012, 05:47:09 am »

I interpret it as in camera HDR but is it just jpeg or RAW HDR?

Hi Marc,

There are two modes, one is HDR and it produces JPEG output, and the other is multiple exposure mode which can output a single Raw and/or JPEG with e.g. the average of up to 9 exposures (1.58 bits noise improvement). The other multiple exposure mode methods seem more creative in the sense that the exposure time for each sub-exposure is reduced and different things can be overlayed without clipping highlights.

We'll have to wait and see how these things perform in real life.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

DeeJay

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 250
Re: Canon 5D III on chip noise reduction?
« Reply #7 on: March 03, 2012, 06:03:56 am »

Sounds more like it is for colour accuracy and reduction of ca, moire and similar effects.
Logged

Nigel Johnson

  • Full Member
  • ***
  • Offline Offline
  • Posts: 126
Re: Canon 5D III on chip noise reduction?
« Reply #8 on: March 03, 2012, 07:50:43 am »

I find it interesting that I have seen no examples of the HDR capability posted. I would think they would be pushing that pretty hard.

Some HDR photos have been posted:

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/02_cinc_big.jpg
Exposure setting: HDR Art Vivid

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/06_cinc_big.jpg
Exposure setting: HDR Art Standard

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/14_cinc_big.jpg
Exposure setting: HDR Natural

In addition three files using multiple exposure modes have been posted:

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/15_cinc_big.jpg
Exposure setting: Multiple Exposure (Bright)

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/17_cinc_big.jpg
Exposure setting: Multiple Exposure (Dark)

http://cpn.canon-europe.com/files/education/technical/inside_canon_eos_5d_mark_iii/18_cinc_big.jpg
Exposure setting: Multiple exposure

Regards
Nigel

(Edit added multiple exposure links)
« Last Edit: March 03, 2012, 08:35:14 am by Nigel Johnson »
Logged

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Re: Canon 5D III on chip noise reduction?
« Reply #9 on: March 03, 2012, 07:38:43 pm »

A quote from DxO's hands on review:
"Each photodiode is equipped with an underlying transistor that provides initial noise processing at the instant of signal transmission. This means that the EOS 5D MK III can achieve a nominal sensitivity of 25,600 ISO, extendable up to 102,400 ISO in H2 mode. According to Canon, an image shot at 25,600 ISO should be completely usable. (We will certainly not fail to verify this in our tests for noise!"
So there is hope that Canon has reduced read noise and can compete with the Sony sensors
Marc
Logged
Marc McCalmont

NikoJorj

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1082
    • http://nikojorj.free.fr/
Re: Canon 5D III on chip noise reduction?
« Reply #10 on: March 05, 2012, 03:02:21 am »

So there is hope that Canon has reduced read noise and can compete with the Sony sensors
On the other hand, they also talk about "the new array of microlenses gathering more light", which they hyped in every new model since the 20D or so with obviously a minimal effect on dynamic range at low ISO without any improvement on read noise...
But yes, the on-chip noise reduction could be a hint that Canon sensors made a significant progress in this area, I just prefer to be first pessimistic and then well surprised. ;)
Logged
Nicolas from Grenoble
A small gallery

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: Canon 5D III on chip noise reduction?
« Reply #11 on: March 05, 2012, 03:43:48 am »

Since Canon PR seems to mention only medium-to-high ISO improvements: Is it likely that adding "on-chip NR" would only boost small signals (high ISO) before the potentially long and noisy analog interconnect to signal amplifier/AD converter so as to improve SNR at high ISO, not to affect (significantly) IQ at high ISO?

-h
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Re: Canon 5D III on chip noise reduction?
« Reply #12 on: March 05, 2012, 04:06:05 pm »

Is it likely that adding "on-chip NR" would only boost small signals (high ISO) before the potentially long and noisy analog interconnect to signal amplifier/AD converter so as to improve SNR at high ISO, not to affect (significantly) IQ at high ISO?
I think that is likely, in part because a Canon publication describing a prototype sensor indicates that is can apply a variable degree of charge amplification in the transfer of signal from sensel to a sense capacitor at the edge of the sensor. This implies that the maximum possible amplification would be too much for use at low ISO speeds with full wells (overloading the charge of voltage limits of circuitry down-stream), so that the higher amplification levels are only used at higher ISO speeds, to help lift the weak signal from less than fully exposed senses further above the noise levels further down the signal path.

I had guessed that Canon was already doing this, and some noise measurements suggest that it is, but maybe it is only arrivi g now in production sensors, or maybe the event of the variable amplification has been increased, to produce even more benefit at the new very hi ISO speeds being offered.
Logged

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Re: Canon 5D III on chip noise reduction?
« Reply #13 on: March 05, 2012, 05:42:58 pm »

I think that is likely, in part because a Canon publication describing a prototype sensor indicates that is can apply a variable degree of charge amplification in the transfer of signal from sensel to a sense capacitor at the edge of the sensor. This implies that the maximum possible amplification would be too much for use at low ISO speeds with full wells (overloading the charge of voltage limits of circuitry down-stream), so that the higher amplification levels are only used at higher ISO speeds, to help lift the weak signal from less than fully exposed senses further above the noise levels further down the signal path.

I had guessed that Canon was already doing this, and some noise measurements suggest that it is, but maybe it is only arrivi g now in production sensors, or maybe the event of the variable amplification has been increased, to produce even more benefit at the new very hi ISO speeds being offered.

Wouldn't read noise include any noise between the sensel and the ADC (along with the accuracy of your ADC)? And isn't the increased DR in the Sony sensors because they have reduced read noise (shorter signal path from sensel to ADC)? So wouldn't amplifying the sensel signal to the (off chip) ADC increase your signal to noise ratio at any ISO thus reducing read noise for all ISO's? Thus improving DR at base ISO? Just hoping for the magic 14 stops of DxO DR!

Marc
Logged
Marc McCalmont

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Re: Canon 5D III on chip noise reduction?
« Reply #14 on: March 05, 2012, 07:58:09 pm »

So wouldn't amplifying the sensel signal to the (off chip) ADC increase your signal to noise ratio at any ISO thus reducing read noise for all ISO's? Thus improving DR at base ISO? Just hoping for the magic 14 stops of DxO DR.
It would, if the same degree of amplification could be applied at any ISO speed. But my point was that if Canon is now using or extending the strategy (described by Canon in a research publication) of applying more amplification at higher ISO speeds than at lower ones, the change compared to previous models might only involve greater amplification than before at high ISO speeds, to bring those signals up nearer to the maximum that the rest of the signal path can handle, while at lower ISO speeds, the signal was already getting as much amplification as could be handled, and so is not getting any more now, so that the lower ISO speeds see no improvement from this change, and the dynamic range at low ISO speeds is not affected.

Of course this does not rule out other improvements, like reduced noise in the sensels themselves, or upgrading the whole signal path from sensel to ADC output to sustain a higher signal to noise ratio like 8000:1 or 16000:1, as neede to get 13 or 14 significant bits in the digital output.
Logged

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: Canon 5D III on chip noise reduction?
« Reply #15 on: March 06, 2012, 02:39:38 am »

If there is significant noise contribution between the sensel and ADC, doing amplification as early as possible would be beneficial. An early amplifier could (as you seem to be suggesting) work as some kind of "AGC", amplifying presumed weak signals more than presumed strong ones. In effect, we would then have something ala:
sensel->g1->add noise->g2->ADC

If desired total gain, g is 2^3=8 compared to base ISO, then increasing g1 from 1.0 to 4.0 would mean decreasing g2 from 8.0 to 2.0 to maintain total amplification. But the added noise component would be multiplied by 2.0 instead of 8.0 (a good thing). The ADC would presumably see the same peak levels (I guess aligned with sensel saturation?), so if g1 has to be limited for low ISO, this means that the analog stages before ADC (g2) allows limited input level before reaching saturation?

In all fairness, I think that more users wish for "improved IQ at high ISO" than "improved DR at low ISO". _If_ Canon are able to significantly improve IQ at high ISO (compared to 5Dmk2 and D800) without compromising other aspects or increasing cost, I am guessing that they will sell many units.

-h
« Last Edit: March 06, 2012, 02:41:15 am by hjulenissen »
Logged

marcmccalmont

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1780
Re: Canon 5D III on chip noise reduction?
« Reply #16 on: March 06, 2012, 03:44:22 am »



In all fairness, I think that more users wish for "improved IQ at high ISO" than "improved DR at low ISO". _If_ Canon are able to significantly improve IQ at high ISO (compared to 5Dmk2 and D800) without compromising other aspects or increasing cost, I am guessing that they will sell many units.

-h
I have the opposite preference base ISO DR is more important to me than High ISO noise
Marc
Logged
Marc McCalmont

hjulenissen

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 2051
Re: Canon 5D III on chip noise reduction?
« Reply #17 on: March 06, 2012, 04:39:55 am »

I have the opposite preference base ISO DR is more important to me than High ISO noise
Marc
So do most of the people in this forum it seems, but I still think that is a vanishingly small minority of the masses that buy cameras like the 5D series.

<gross simplification>
Most people take pictures of their dogs and children. They use in-camera jpeg and have never heard of "curves and levels". Most people use their iPhone as their main camera. If only a small fraction of those have the funds and inclination to buy the 5Dmk3, they will probably outnumber the landscape photographers by a large factor. If those people did not buy cameras like the 5Dmk3, the landscape photographers would probably have to pay closer to MFDB-prices to get a digital FF camera with MFDB ergonomy.
</gross simplification>

I would like to have both. I think that having a large DR at base ISO is a good starting-point to do well at high ISO as well, but I dont know enough about the physical/economical trade-offs to say this with confidence.

-h
« Last Edit: March 06, 2012, 04:46:14 am by hjulenissen »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Re: Canon 5D III on chip noise reduction?
« Reply #18 on: March 06, 2012, 09:00:29 am »

I have the opposite preference base ISO DR is more important to me than High ISO noise
Marc
To put this in context, the SNR is always better at base ISO speed than at higher speeds; this approach of more early amplification of the signal at higher ISO speeds is just a way to reduce the SNR disadvantage in higher speed shooting. That is, it is more or less independent of anything that Canon could also be doing to improve SNR at low speeds.

On that latter goal, I see only two likely scenarios:

1. The two new Canon 35mm format bodies offer better pixel level SNR at base ISO speed (so called dynamic range) than what Sony and Nikon have been offering in 24MP sensors for several years, and better than with D800 output down-sampled to match the Canon bodies for resolution.

2. Canon has fallen fundamentally behind its main competition in sensor technology for still photography with decent light levels, and the only way to overcome that is something akin to the Sony approach of doing ADC on-chip, closer to to the sensel and probably with more ADC units to allow each ADC to operate at a lower rate.

In case 2, it is possible I suppose that Canon simply does not prioritize extremes of dynamic range, which are only visible when the image is displayed with a heavy artificial reduction of the brightness range between shadows and highlights. This could be because Canon has decided that using multiple frames and HDR is an adequate solution for those situations, or because Canon has decided that there is better ROI to be had from putting R&D resources into being the leader in DSLR video performance.
« Last Edit: March 06, 2012, 09:03:00 am by BJL »
Logged

BJL

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 6600
Re: Canon 5D III on chip noise reduction?
« Reply #19 on: March 06, 2012, 10:25:30 am »

I think that having a large DR at base ISO is a good starting-point to do well at high ISO as well, but I dont know enough about the physical/economical trade-offs to say this with confidence.
True up to a point, but with some complications. It seems clear that reducing the noise level within the sensel (while keeping full well capacity the same) increases the SNR of the signal within the sensel, which can certainly help with both improving DR at base ISO speed and reducing noise levels (improving SNR) at higher ISO speeds.

However, the catch with recent Canon sensisthmuses to be that the signal processing and transmission path beyond the sensel has a lower SNR than what the sensel itself has. If I recall correctly, measurements cited by Emil Martinec suggest that in one high end Canon sensor, the signal path consisting of charge to voltage conversion, variable amplification for adjusting ISO speed, and analog-to-digital conversion has only an 8000:1 SNR (about 13 stops, or "13 bits worth") even though the sensel has a roughly 20000:1 ratio between full well capacity and its noise floor. In that sort of scenario, when the exposure is a few stops faster than base ISO, so that the highest electron count at any sensel is 1/4 or less of full well capacity, then the ratio of maximum sensel signal to noise level is below the 8000:1 that the rest of the system can handle. Then, high ISO performance can be improved up to the 8000:1 limit with changes in the level amplification, but low ISO performance needs a quite different change, improving on that 8000:1 downstream of the sensel.
Logged
Pages: [1] 2   Go Up