Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
03-09-2011, 04:41 PM   #46
Pentaxian
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,862
QuoteOriginally posted by Smeggypants Quote
for example if you've got a 14bit convertor and your brightest highlight is, say, 30db down ( 5 stops I think ) then you are only digitising the pic with a 9 bit dynamic range
This sounds like a simple and straight argument.

Still it can break if you take the inevitable photon shot noise into account (that's the noise in the photon stream before it enters the lens). And for the K-5, it does break. The exact numbers are in my posts higher up.

03-10-2011, 08:56 AM   #47
Senior Member




Join Date: Nov 2006
Location: Oaxaca, Mexico; Shohola PA, USA
Posts: 219
QuoteOriginally posted by dosdan Quote
...whereas with high ISO in-camera you can blow the highlights when you're trying to expose the histogram to the right to maximise noise performance.
I do not understand this. If, as Falk wrote, "ISO isn't (and has never been) a fundamental exposure parameter", how is increasing ISO "exposing to the right"? Isn't ETTR only valid at base ISO, since increasing ISO does not increase the amount of light received by the sensor? Doesn't increasing ISO just gives the illusion of ETTR, because it moves the histogram to the right?
QuoteOriginally posted by dosdan Quote
With the K5 you can expose to the centre or more to the left without much of a noise penalty at all.
Is there any penalty from exposing to the left? Put another way, is there a benefit from maximizing exposure, short of clipping at base ISO? I do not have a technical background, but common sense (which is not always right, of course) tells me that giving the sensor more photons to work with is always going to be better than giving it fewer.

Jeff
03-10-2011, 09:23 AM   #48
Senior Member




Join Date: Feb 2011
Location: Tallinn
Photos: Gallery
Posts: 162
While all this is interesting reading I can not seem to understand the practical benefit of it. Does it mean that when shooting in ISO range 100-1600 you could get the same results by correcting exposure in PP. Like if I have picture rather dark at 100 I can make it look as good in PP as if shot 1600 well exposed?
Obviously there is much more *grain* with higher ISO.
03-10-2011, 01:12 PM   #49
Pentaxian
dosdan's Avatar

Join Date: Dec 2007
Location: Brisbane, Australia
Photos: Gallery | Albums
Posts: 1,697
Original Poster
QuoteOriginally posted by Jeff Charles Quote
Isn't ETTR only valid at base ISO, since increasing ISO does not increase the amount of light received by the sensor? Doesn't increasing ISO just gives the illusion of ETTR, because it moves the histogram to the right?

Is there any penalty from exposing to the left? Put another way, is there a benefit from maximizing exposure, short of clipping at base ISO? I do not have a technical background, but common sense (which is not always right, of course) tells me that giving the sensor more photons to work with is always going to be better than giving it fewer.

ETTR at any ISO setting where analogue amplification is used (not above ISO1600) will also give a bigger signal to the ADC. This is particularly important if the ADC is noisy. The K-5s ADC is not.

"'Giving the sensor more photons" means either exposing longer or with a wider aperture - that's a good idea, but not always possible. Changing ISO sensitivity on a DSLR is amplifying that light, after capture and conversion to an electrical signal, either in the analogue or digital realm.

I'm not recommending ETTL per se. Rather, there just isn't a strong imperative to ETTR with this model. If you're shooting a tricky scene for exposure, there's hardly any noise penalty leaving it somewhat under-exposed, rather than worrying about "I must always ETTR".

This camera does receive some benefit from boosting the in-camera ISO up as far as ISO1600, or perhaps ISO1100 or ISO1250 (depending on the size you've set for ISO steps) if you've enabled the ISO80 extension, but it's not much compared to other models. As other sensors match or exceed its read noise performance, fixed ISO operation becomes more feasible, but it would be most convenient when both hardware & software fully support it.

Dan.


Last edited by dosdan; 03-19-2011 at 01:16 AM.
03-10-2011, 02:46 PM   #50
Senior Member




Join Date: Nov 2006
Location: Oaxaca, Mexico; Shohola PA, USA
Posts: 219
QuoteOriginally posted by dosdan Quote
...

I'm not recommending ETTL per se. There just isn't a strong imperative to ETTR with this model. If you're shooting a tricky scene for exposure, there's hardly any noise penalty leaving it somewhat under-exposed, rather than worrying about "I must always ETTR".
Thanks. That makes sense. The K-5 tolerates a little under exposure without much penalty. In practical terms, that could mean using a faster shutter speed to be sure to freeze the action. With a camera less tolerant of underexposure, I might try to "get away with" a slower shutter speed to avoid underexposure and more noise.
03-10-2011, 03:20 PM   #51
Veteran Member




Join Date: Sep 2007
Photos: Gallery
Posts: 975
very well summed up, i was wondering when somebody will put all this together (after all the "turning black picture to normal picture from a k-5" stories).

iirc, there used to be some "auto adjustment" feature on pentax dslrs, for the jpegs. if one would enable that, the preview in the raw file should not show dark anymore when using constant iso and underexposing. (i can't try to find it now, maybe somebody can)

interestingly enough, this makes fancy metering systems on dslrs look very silly now, doesn't it? you really only need a ballpark figure now, to have an idea where your optimum snr would be, and from there you can adjust to chose your compromises (based on desired dof and shutterspeed ranges)

audio/foto analogy: i think that breaks when you go in too deep, because audio is essentially analogue, while light is essentially digital (in the quantum sense of the word), or in other words, light is not continuous when you try to "count" it, so it is by nature "digital", while sound is a continuous wave, so by nature analogue. which is very strange, come to think of it, and might even be arguable . in theory, if one could make such a device, you should not need an adc for light sensors: such a device meaning a foton counter which outputs a number somehow, not an analogue signal (voltage), currently we're basically doing dac-adc, when measuring light

Last edited by nanok; 03-10-2011 at 03:25 PM. Reason: 1k typos
03-10-2011, 04:41 PM   #52
Veteran Member
Smeggypants's Avatar

Join Date: Sep 2010
Posts: 1,536
QuoteOriginally posted by dosdan Quote
ETTR at any ISO setting where analogue amplification is used (not above ISO1600) will also give a bigger signal to the ADC. This is particularly important if the ADC is noisy. The K-5s ADC is not.

Notwithstanding Falk's post above, a 'bigger' signal to the ADC will mean the signal is digitised with a higher bit resolution. This can be important for keep rounding errors down when performing multiple PP operations. In that respect it's better to start with a signal that is as higher resolution as possible.
03-10-2011, 05:00 PM   #53
Pentaxian
dosdan's Avatar

Join Date: Dec 2007
Location: Brisbane, Australia
Photos: Gallery | Albums
Posts: 1,697
Original Poster
QuoteOriginally posted by Smeggypants Quote
Notwithstanding Falk's post above, a 'bigger' signal to the ADC will mean the signal is digitised with a higher bit resolution. This can be important for keep rounding errors down when performing multiple PP operations.
Not sure if "rounding error" is the right term here. This is a dithered signal. The min. read noise level is 3.3 e- to 2.2 e-, depending on ISO, according to Falk, so this above the LSB. Anyway, PP processing is usually done at much higher internal precision than 14 bits. I think the only thing being rounded, if at all, is the representation of the noise.

Dan.

03-10-2011, 05:09 PM   #54
Veteran Member
Smeggypants's Avatar

Join Date: Sep 2010
Posts: 1,536
QuoteOriginally posted by dosdan Quote
Not sure if "rounding error" is the right term here. This is a dithered signal. The min. read noise level is 3.3 e- to 2.2 e-, depending on ISO, according to Falk, so this above the LSB. Anyway, PP processing is usually done at much higher internal precision than 14 bits. I think the only thing being rounded, if at all, is the representation of the noise.

Dan.
It doesn't matter if the signal was previously dithered or not.

When you perform DSP ( i,e perform mulitplication or division on a digital signal ) you get rounding errors. These errors are cumulative the more DSP operations you perform on the signal and manifest themselves as increased distortion.

Further huge internal prescision ( Like 80bit intel registers for example ) makes little difference of the prescision of the original signal is only, say, 6 or 7 bits ( or less ) becuase you exposed to the left and didn't give the ADC an optimal signal
03-10-2011, 06:01 PM   #55
Pentaxian
dosdan's Avatar

Join Date: Dec 2007
Location: Brisbane, Australia
Photos: Gallery | Albums
Posts: 1,697
Original Poster
QuoteOriginally posted by Smeggypants Quote
It doesn't matter if the signal was previously dithered or not.

When you perform DSP ( i,e perform mulitplication or division on a digital signal ) you get rounding errors. These errors are cumulative the more DSP operations you perform on the signal and manifest themselves as increased distortion.

Further huge internal prescision ( Like 80bit intel registers for example ) makes little difference of the prescision of the original signal is only, say, 6 or 7 bits ( or less ) becuase you exposed to the left and didn't give the ADC an optimal signal
But the rounding err in this case will only be significant on the first ADC operation.

Say, you have a quantization step of 1 i.e. 2->3, Then 2.6 would initially be rounded up to 3 in say a 14-bit range of 0-16383. From then on, as long as the internal precision was high enough, all further rounding errors will not be significant.

Dan.

Last edited by dosdan; 10-21-2011 at 03:38 AM.
03-10-2011, 06:48 PM   #56
Veteran Member
Smeggypants's Avatar

Join Date: Sep 2010
Posts: 1,536
QuoteOriginally posted by dosdan Quote
But the round err in this case will only be significant on the first ADC operation.

Say, you have a quantization step of 1 i.e. 2->3, Then 2.6 would initially be rounded up to 3 in say a 14-bit range of 0-16383. From than on, as long as the precision was high enough, all further rounding errors will not be significant.

Dan.
If you take an analogue signal and digitise it with, say 14 bit precision then that signal will be a more accurate representation than than it would have been if you digitise it with, say, 5 bit precision.

The 14bit version will also be more robust to subsequent DSP performed upon it
03-10-2011, 08:42 PM   #57
Pentaxian
dosdan's Avatar

Join Date: Dec 2007
Location: Brisbane, Australia
Photos: Gallery | Albums
Posts: 1,697
Original Poster
QuoteOriginally posted by Smeggypants Quote
If you take an analogue signal and digitise it with, say 14 bit precision then that signal will be a more accurate representation than than it would have been if you digitise it with, say, 5 bit precision.

The 14bit version will also be more robust to subsequent DSP performed upon it
In this discussion, we've not been taking about a 9-bit drop in precision.

Compared to a FS quantized signal at each of the full-stop ISO settings, obtained through analogue amplification, the loss in precision if leaving the camera at ISO100 is:

ISO100 - no loss
ISO200 - 1 bit loss
ISO400 - 2 bits loss
ISO800 - 3 bits loss
ISO1600 - 4 bit loss

Above that, it's digital multiplication anyway, so there's no difference.

Say you had a full 14 bits precision at ISO100. If the light then dropped 2 stops and you choose to continue to shot at ISO100 instead of ISO400, the precision would drop to effectively 12 bits after you boosted it 2 stops in PP.

4 bits loss of precision sounds a lot, but I doubt you'd notice it in an equivalent-to-ISO1600 image. Go to http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#bitdepth and mouse over Figs 17 & 20 where you can see no apparent change in an adequately dithered image after major bit-depth reductions.

So it's a question of would there be much further losses due to multiple rounding error in the processing of a 10 bits effective precision signal at ISO1600. I don't think there would be, as long as the internal precision was high.

Dan.

Last edited by dosdan; 03-10-2011 at 09:19 PM.
03-11-2011, 04:34 PM   #58
Pentaxian
dosdan's Avatar

Join Date: Dec 2007
Location: Brisbane, Australia
Photos: Gallery | Albums
Posts: 1,697
Original Poster
Some more info from Guillermo Luijk (translated) regarding bit-depth vs ISO:

And the same will happen with any camera: only the lowest ISO setting will be the provision to require the maximum number of bits of A / D converter. As we increase the ISO, the noise increase will be redundant to pass one by one the least significant bits of RAW.

Google Translate

Mouseover Fig. 9. Admittedly, this is a high noise camera, but there is no effective quality difference between 12-bit & 8-bit in the test image at ISO1600.

Dan.
03-19-2011, 07:37 PM   #59
Pentaxian
dosdan's Avatar

Join Date: Dec 2007
Location: Brisbane, Australia
Photos: Gallery | Albums
Posts: 1,697
Original Poster
Guillermo posted an example of a wide DR shoot taken with a K-5 at ISO100 and pushed 4 stops in PP, compared to the same shot taken at ISO1600.

Noise and DR comparision: Canon 5D vs 5D2 vs 7D vs Pentax K5

His conclusions:

  • In the highlights the 4 extra stops allow to properly capture the lamp area and view through the window at ISO100, not at ISO1600

  • In the medium shadows, no SNR improvement for pushing ISO

  • In the deep shadows, it seems read noise becomes important and a bit more noise can be seen in the ISO100 image. Also many pixels get clipped to 0 in the RAW file (perhaps the 14-bit RAW encoding was not enough here) producing darker areas lacking detail that displays better in the ISO1600 shot

Dan.
03-19-2011, 08:14 PM   #60
Veteran Member
Smeggypants's Avatar

Join Date: Sep 2010
Posts: 1,536
I don't have a problem using the ISO parameter. I don't think anyone else really does in practise. I'm quite happy for the camera to provide some analogue amplification before the converters rather than do it afterwards.

I can have an ISOless camera now if I want to shoot manual and ignore the meter. It's not going to produce better pictures though so why bother?

Opnce you get into PP cumulative DSP operations will have less of a degrading effect when the original signal is converted to digital with more an increased bit accuracy.

I think you are evangelising over something that while might be theoretically possible in the future, it's pretty pointless.


Have you tried shooting ISOLess already? if not? why not?
Reply

Bookmarks
  • Submit Thread to Facebook Facebook
  • Submit Thread to Twitter Twitter
  • Submit Thread to Digg Digg
Tags - Make this thread easier to find by adding keywords to it!
camera, cameras, dslr, ev, iso, k-5, k-5 ii, k-5 iis, k5, noise, pentax k-5, sensor
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
Post your "Park Bench" "or "Picnic Table" images tessfully Mini-Challenges, Games, and Photo Stories 1347 1 Day Ago 12:52 PM
Please explain "Sensor is Outresolving Some Lenses" Parallax Pentax K-5 21 11-24-2010 05:05 PM
Sensor shift during "long" exposures eclipsed450 Pentax DSLR Discussion 11 12-18-2009 12:51 PM
K10D "Sensor moving" possible problem? jaitas Troubleshooting and Beginner Help 6 05-07-2009 02:08 PM
"Green Clean" Sensor Cleaning System. NLAlston Pentax Camera and Field Accessories 2 05-16-2007 07:35 PM



All times are GMT -7. The time now is 11:21 PM. | See also: NikonForums.com, part of our network of photo forums!
  • Red (Default)
  • Green
  • Gray
  • Dark
  • Dark Yellow
  • Dark Blue
  • Old Red
  • Old Green
  • Old Gray
  • Dial-Up Style
Hello! It's great to see you back on the forum! Have you considered joining the community?
register
Creating a FREE ACCOUNT takes under a minute, removes ads, and lets you post! [Dismiss]
Top