Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
11-06-2007, 08:53 PM   #16
Inactive Account




Join Date: Mar 2007
Location: CA
Posts: 24
I'd have to say this is a most useful post on the K10d. It is very much validated. It's counterintuitive, but I often get better shots at ISO 800 than ISO 400 due to the combination of exposure and shutter speed to minimize motion blur. Add SR and it's a killer combo!

Push processing is still useful but exacting white balance and manual metering can help make that more successful.

11-06-2007, 09:02 PM   #17
Veteran Member
distudio's Avatar

Join Date: Jul 2007
Location: Sydney
Photos: Albums
Posts: 448
Two things that make ISO quite acceptable (IMO) in the instance cited are that firstly the light is pretty well balanced (set up for TV broadcast) so the colour channel gain will be more like daylight not at extremes with one channel driving into the noise such as is often the case in poor light.

Secondly since there is a lot of light at these types of venues I expect that the high ISO has been selected primarily to provide relatively high shutter speeds to suspend motion not in a to fight to get a barely hand-holdable shutter speed so noise will be less of a problem than in more lengthy exposures.

Take some ISO 1600 shots at slow shutter speeds using a pair of badly placed 150W green and red overhead para-floods in a dingy pub with the WB cranked to get back some semblance of colour and then see how the noise is ;-)

Whether the K10D high ISO noise is acceptable or not is up to the individual, often I find it quite poor.
11-07-2007, 12:47 AM   #18
Junior Member




Join Date: Mar 2007
Location: Oregon, USA
Photos: Gallery
Posts: 49
QuoteOriginally posted by Ray Pulley Quote
More importantly lets think of the 0-255 range in stops of light:

If you expose to the right and exactly fill the histo from black to pure white, one stop from the top is 128 steps because one stop from the top (255) is 1/2 as much light (the highlights).

Subtract another stop (minus 1/2 of the light) and you get: 128 - 64 = 64, so this stop is only 64 steps.

...
I'm pretty sure the 0-255 range doesn't map linearly to number of photons--it's a higher order mapping. So 128 is a lot more than 1 stop below 255--probably more like 2 or 3 stops.

The RAW image (12 bits) might be linear.

QuoteQuote:
In terms of noise, the issue is simply the number of photons captured by the pixel well versus the absolute noise that cannot be reduced below a certain level. Exposing to the right simply means getting as many photons into the pixels as possible for a given scene, which maximizes the signal to noise ratio (more good photons in the pixel while the number of noise photons remain the same). If the max number of photons is represented by 255 and the noise equals 1 (it is far smaller than this), then exposing to the right gives you a signal to noise ratio of 255:1, while exposing to 128 gives you 128:1, making the noise look 2X as bad.
It's a little more complicated than that. SNR due to photon quantization isn't varying linearly with number of photons. It improves as the square root of photon count. There's also readout noise portion of SNR--that does go down directly with number of photons. The total improvement of noise will be somewhere between the two. As you get closer to the shadow end, readout noise becomes more significant and so you might observe SNR in the shadows almost increasing linearly with number of photons.

Doesn't change the correctness of your conclusion--ie., for lowest SNR, the farther to right you expose (without blowing highlights), the better.

QuoteQuote:
I think that the idea behind using 1600 and exposing to the right versus underexposing at 800, is that the added amplification noise of the camera at 1600 ISO is not as bad as the reduction in signal to noise ratio that comes with one stop underexposure at 800 which then has to be bumped up one stop in software during PP. You can test this yourself by shooting the exact same scene exposed to the right at ISO 1600 and then one stop underexposed at 800. Open both images in your editing software and push the exposure up one stop for the ISO 800 shot and compare the noise.
Are you sure the camera even uses analog amplification to achieve higher ISO? At least on the K10d, I'd think with a 22-bit ADC and 16-bit datapath, there'd be no need for analog amplification.

Anyway, if it does use analog amplification, then you're right, you should theoretically get better SNR at higher ISO for equal exposure settings (f-number and shutter speed)--although the difference may be negligible if the noise from the sensor is already much larger than the ADC noise or other noise sources after the sensor. If it doesn't use analog amplification, then there will be no difference.

In my experience with the K10d, ISO1600 EV0 is less noisy than ISO800 EV-1, but it's also pretty obvious that in-camera NR has been applied--even to the RAW data. If you run both images through auto-profiling NR software to put everything back on a level playing field (eg., Noiseware), they come out looking virtually identical--both in terms of noise and in terms of detail retention.

Bart
11-07-2007, 06:12 AM   #19
Pentaxian
Lowell Goudge's Avatar

Join Date: Jan 2007
Location: Toronto
Photos: Gallery | Albums
Posts: 15,400
QuoteOriginally posted by bart_hickman Quote
I'm pretty sure the 0-255 range doesn't map linearly to number of photons--it's a higher order mapping. So 128 is a lot more than 1 stop below 255--probably more like 2 or 3 stops.

The RAW image (12 bits) might be linear.
I still need to check this at iso 1600 but at ISO 800 the K10D has the following characteristics with JPEG images.

Between 1 and 25 (histogram level) there are 1-2 stops of data compressed non linearly
The same is true at the high end between 230 and 255.
The middle region has between 4 and 6 stops linearly spaced, i.e. between about 30 and 50 to a stop (as a function of contrast selected in the camera).

I have not checked RAW

The graph shown is a surprise as it is linear through the maximum, which will have a hard transition, not a gradual one. Personally I think it is wrong.

BTW. For anyone interested in checking this for themselves, the test is quite simple.

Meter at maximum apature in manual mode, and take a picture at each F stop (without changing shutter) to minimum apature.
Repeat this metering at minimum apature and take a shot at each stop towards maximum.

If you plot grey scale vs stops -ve for the first set and +ve for second set you should get a graph like the one shown.

The best target is a paved road or block (concrete wall)

11-07-2007, 10:03 AM   #20
Veteran Member




Join Date: Jan 2007
Location: Northern California
Posts: 475
QuoteOriginally posted by bart_hickman Quote
I'm pretty sure the 0-255 range doesn't map linearly to number of photons--it's a higher order mapping. So 128 is a lot more than 1 stop below 255--probably more like 2 or 3 stops.

The RAW image (12 bits) might be linear.



It's a little more complicated than that. SNR due to photon quantization isn't varying linearly with number of photons. It improves as the square root of photon count. There's also readout noise portion of SNR--that does go down directly with number of photons. The total improvement of noise will be somewhere between the two. As you get closer to the shadow end, readout noise becomes more significant and so you might observe SNR in the shadows almost increasing linearly with number of photons.

Doesn't change the correctness of your conclusion--ie., for lowest SNR, the farther to right you expose (without blowing highlights), the better.



Are you sure the camera even uses analog amplification to achieve higher ISO? At least on the K10d, I'd think with a 22-bit ADC and 16-bit datapath, there'd be no need for analog amplification.

Anyway, if it does use analog amplification, then you're right, you should theoretically get better SNR at higher ISO for equal exposure settings (f-number and shutter speed)--although the difference may be negligible if the noise from the sensor is already much larger than the ADC noise or other noise sources after the sensor. If it doesn't use analog amplification, then there will be no difference.

In my experience with the K10d, ISO1600 EV0 is less noisy than ISO800 EV-1, but it's also pretty obvious that in-camera NR has been applied--even to the RAW data. If you run both images through auto-profiling NR software to put everything back on a level playing field (eg., Noiseware), they come out looking virtually identical--both in terms of noise and in terms of detail retention.

Bart
Bart, no argument and thanks for the additional detail.

Yes, I was over-simplifying for the purposes of clarification of the concept. I thought I was clear about that, but maybe not.

In practice, if you keep the concept in mind and apply it as often as possible, you will get images with the best possible use of the DR of the camera regardless of the technical reasoning behind it. I use this rule of thumb along with a modified zone system quite often where I spot on white and add 1.5 stops to the exposure and then chimp to see if I am as far right as I can get without going over. I shoot only RAW and this works pretty well for me, assuming that there is any white in my scene Black works just as well, but in an inverted way (subtract 1.5 or so stops).

You are correct about the noise behavior as well, but I think that the person I was responding to had already stated that the technical reasoning of some of these rules of thumb was not something he understood or perhaps not something he wanted to understand that much. Frankly many folks cannot, or do not want to, apply such specific technical information to their regular use of the camera and therefore I do not think such details are all that helpful. This is not a slam against any of those folks, as it is not necessary to understand any of the specific details to get good images. After all, how much did we know (or care) about how film emulsion was made and how every layer worked and responded to light? Most of us just came to understand how the film responded to certain situations and then adpated our techniques to get the final image we wanted.

Are the details correct? Perhaps, assuming that they are not based upon guesses about things we do not know for sure, but they are sometimes hard to translate into taking pictures and specific results. Expose to the right is much easier to understand and you can see the results directly when PP'ing the images.

Many parts of the signal path in the K10D (perhaps almost all) run directly through the NuCore chip, which has many features that Pentax may or may not have chosen to use/implement, so it is not easy to decide exactly what is going on, and honestly, it really does not matter as it is not like we can change it. It is interesting to talk about, however.

I do not think anyone has figured out for sure if there are analog amplification steps in the K10D system design. However, this propaganda document form NuCore would lead me to believe that the gain is adjusted in an analog section which is handled by the chipset:

http://www.nucoretech.com/nu3/images/80_downloads/pres_technology.ppt.us.pdf

Lastly, I purposely did not state that I agreed with the "shoot higher ISO to the right" versus shooting a stop lower as I have not really made up my mind whether there would really be any noticeable difference in doing this or not. I have also not taken the time to test the idea on the K10D as I do not recall ever shooting that camera above 400 ISO. Things just get too ugly for my taste above that, so I reach for the K100D in those situations.

Thanks again for the added technical bits.

Ray
11-07-2007, 10:55 AM   #21
Veteran Member




Join Date: Jan 2007
Location: Northern California
Posts: 475
QuoteOriginally posted by bart_hickman Quote
I'm pretty sure the 0-255 range doesn't map linearly to number of photons--it's a higher order mapping. So 128 is a lot more than 1 stop below 255--probably more like 2 or 3 stops.

The RAW image (12 bits) might be linear.



It's a little more complicated than that. SNR due to photon quantization isn't varying linearly with number of photons. It improves as the square root of photon count. There's also readout noise portion of SNR--that does go down directly with number of photons. The total improvement of noise will be somewhere between the two. As you get closer to the shadow end, readout noise becomes more significant and so you might observe SNR in the shadows almost increasing linearly with number of photons.

Doesn't change the correctness of your conclusion--ie., for lowest SNR, the farther to right you expose (without blowing highlights), the better.



Are you sure the camera even uses analog amplification to achieve higher ISO? At least on the K10d, I'd think with a 22-bit ADC and 16-bit datapath, there'd be no need for analog amplification.

Anyway, if it does use analog amplification, then you're right, you should theoretically get better SNR at higher ISO for equal exposure settings (f-number and shutter speed)--although the difference may be negligible if the noise from the sensor is already much larger than the ADC noise or other noise sources after the sensor. If it doesn't use analog amplification, then there will be no difference.

In my experience with the K10d, ISO1600 EV0 is less noisy than ISO800 EV-1, but it's also pretty obvious that in-camera NR has been applied--even to the RAW data. If you run both images through auto-profiling NR software to put everything back on a level playing field (eg., Noiseware), they come out looking virtually identical--both in terms of noise and in terms of detail retention.

Bart
Bart,

One more note:

Regardless of how the scene DR maps to the capture DR (jpeg, or RAW) if you expose such that you use only 1/2 of the range, then you are signifcantly compressing the entire DR of the scene into a much smaller part of the available camera DR, which means that you are limiting the amount of information that will be available for printing or whatever you are doing with the image.

That's the real idea behind "expose to the right".

Ray
11-07-2007, 10:39 PM   #22
Junior Member




Join Date: Mar 2007
Location: Oregon, USA
Photos: Gallery
Posts: 49
QuoteOriginally posted by Ray Pulley Quote
Bart,

One more note:

Regardless of how the scene DR maps to the capture DR (jpeg, or RAW) if you expose such that you use only 1/2 of the range, then you are signifcantly compressing the entire DR of the scene into a much smaller part of the available camera DR, which means that you are limiting the amount of information that will be available for printing or whatever you are doing with the image.

That's the real idea behind "expose to the right".

Ray
Hi Ray,

Yes, I agree. Aside from technical curiosity, I was pointing it out because the consequences of underexposing by, say, 1 stop aren't nearly as serious as they would be if the jpeg mapping were linear.

Thanks for posting that NuCore spec. It looks like they can apply gamma correction right after the A/D, so that means the K10d RAW files are probably even non-linear (which is a good thing.)

Bart

Last edited by bart_hickman; 11-07-2007 at 10:49 PM.
11-07-2007, 10:45 PM   #23
Junior Member




Join Date: Mar 2007
Location: Oregon, USA
Photos: Gallery
Posts: 49
QuoteOriginally posted by Lowell Goudge Quote
I still need to check this at iso 1600 but at ISO 800 the K10D has the following characteristics with JPEG images.

Between 1 and 25 (histogram level) there are 1-2 stops of data compressed non linearly
The same is true at the high end between 230 and 255.
The middle region has between 4 and 6 stops linearly spaced, i.e. between about 30 and 50 to a stop (as a function of contrast selected in the camera).
Hi Lowell,

Just remember that having a linear response versus stops isn't actually linear--it's logarithmic. Ie., the number of photons that gets from 3EV to 4EV is twice as many as it took to get from 2EV to 3EV.

Bart

11-08-2007, 05:42 AM   #24
Pentaxian
Lowell Goudge's Avatar

Join Date: Jan 2007
Location: Toronto
Photos: Gallery | Albums
Posts: 15,400
QuoteOriginally posted by bart_hickman Quote
Hi Lowell,

Just remember that having a linear response versus stops isn't actually linear--it's logarithmic. Ie., the number of photons that gets from 3EV to 4EV is twice as many as it took to get from 2EV to 3EV.

Bart
I am not talking about linearity of the sensor to light, what I am discussing is how photo editing programs, and the histogram values correspond to Exposure Vvalue (EV) or F stops.

What I am saying is that the grey scale histogram values are "Linear" with F Stops over the range of about 25 to 230. giving between 4 and 6 stops in this range, depending upon the contrast selected.

At "Nominal" contrast, every measure of 40 in grey scale is 1 stop. This is in reference to the graph someone has shown showing grey scale vs EV. That is all
11-08-2007, 11:52 AM   #25
Veteran Member




Join Date: Jan 2007
Location: Northern California
Posts: 475
QuoteOriginally posted by bart_hickman Quote
Hi Ray,

Yes, I agree. Aside from technical curiosity, I was pointing it out because the consequences of underexposing by, say, 1 stop aren't nearly as serious as they would be if the jpeg mapping were linear.

Thanks for posting that NuCore spec. It looks like they can apply gamma correction right after the A/D, so that means the K10d RAW files are probably even non-linear (which is a good thing.)

Bart
Hi Bart,

It seems like getting detailed information on the newer NuCore chip requires buying the development kit which includes lots of interesting things, but that is the only place I saw a datasheet listed for the chipset used in the K10D.

However, I did find a more complete description of an earlier, less powerful chipset, and if the newer chipset is similar (the block diagrams are similar), it does look like gain is set by a PGA in the chipset and that the jpegs are also likely generated using the chipset. Sharpening and other image tuning are done by the NuCore as well, but seem to all be tunable.

There is an application that you get with the development kit that allows tuning of all of these parameters and probably more, so it seems that Pentax could tune the K10D jpegs if they wanted to. Of course, I do not know how much tuning adjustment range there is available, or if the adjustments that you can do this way would resolve the issues some have with the jpeg output and sharpening methods. I suspect that they could make them look however they wanted to, but as is usual with Pentax, they are stubborn about the design decisions they make even if they turn out to be wrong or unpopular. I am not sure that this is all that different than Nikon and Canon, frankly.

I am pretty sure that some Nikon models used this same chipset or a similar one, as they used to be listed on the NuCore website, but NuCore removed them some time ago. I believe that one such model is the Nikon D2X.

Ray
11-08-2007, 12:01 PM   #26
Veteran Member
mattdm's Avatar

Join Date: Feb 2007
Location: Boston, MA
Photos: Gallery
Posts: 2,964
QuoteOriginally posted by Ray Pulley Quote
There is an application that you get with the development kit that allows tuning of all of these parameters and probably more, so it seems that Pentax could tune the K10D jpegs if they wanted to. Of course, I do not know how much tuning adjustment range there is available, or if the adjustments that you can do this way would resolve the issues some have with the jpeg output and sharpening methods. I suspect that they could make them look however they wanted to, but as is usual with Pentax, they are stubborn about the design decisions they make even if they turn out to be wrong or unpopular.
I haven't done bench tests or anything, but as I understand it, the real solution here is simply for Pentax to separate the sharpening mode from the tone curve -- right now, bright and natural appear to use both different sharpening approaches and different tones. I think the problem would be quite simply addressed by making that two separate menu choices.
11-08-2007, 12:36 PM   #27
Veteran Member




Join Date: Aug 2007
Photos: Gallery
Posts: 1,563
Let's ask Pentax for a new feature

I think this problem has been made very complex, although I've learned a lot with all the responses so far. Thanks everybody.

Suppose we do not apply any noise reduction, than amplifying the sensor signal also amplifies the noise coming in. Regardless if it is done by analog amplification of digital multiplication and offset.

It is very clear that with any picture taken the signal / noise ratio is better in light areas. And that with dark pictures we get a relative low image signal to the noise coming in. All independent of the noise source.

See picture below.

Attachment 5811

If we keep our picture at the right side of the histogram, we maintain a better s/n ratio. However we end up with less DR. I guess nobody will argue that.

This means that the means of the amplification is of no significance. Or is it?
Boosting the signal by increasing ISO value of by increasing the Ev value is a different way of doing the same thing. You could as well shoot at ISO 800 and increase Ev +1 with the SAME shutter speed and apperture. While you have not add any light to the sensor.

Doubling the light by shooting at half the speed in both ISO 1600 or ISO 800 would be beside the point because you are using high ISO values in order to maintain a relative fast shutter speed in the first place.

So, if one can trade in DR for s/n ratio, why not ask Pentax to add this to the feature set of the future K20D? A custom option for the green button in P mode at ISO 1600: half the DR with half the noise?

What do you think of that?

- Bert

Last edited by bymy141; 11-29-2007 at 09:33 AM.
11-08-2007, 01:23 PM   #28
Pentaxian
Lowell Goudge's Avatar

Join Date: Jan 2007
Location: Toronto
Photos: Gallery | Albums
Posts: 15,400
bert

I think there is an additional factor, without going too deep into it,

Changing ISO, as far as I know also changes directly the voltage at the sensor, not just the gain of an amplifier after it.

As a result, it is not just that you are amplifying (or stretching) the dynamic range, and thereby increasing the difference between each pixel value, but by changing the voltage at the chip, you change the excitation for leakage current across the chip, it is the leakage current that causes a lot of the noise. at low ISO this is also chip voltage, and as a result, lower leakage current. It is the same reason that reducing the sensor area (i.e. each individual sensor) has increased noise, the impact of each electron of leakage current is much greater. Heat also makes this noise worse, although I have not seen, and am not suggesting people go out and bake or freeze their cameras to evaluate the impact of noise,
11-08-2007, 02:14 PM   #29
Veteran Member




Join Date: Aug 2007
Photos: Gallery
Posts: 1,563
QuoteOriginally posted by Lowell Goudge Quote
bert

I think there is an additional factor, without going too deep into it,

Changing ISO, as far as I know also changes directly the voltage at the sensor, not just the gain of an amplifier after it.

As a result, it is not just that you are amplifying (or stretching) the dynamic range, and thereby increasing the difference between each pixel value, but by changing the voltage at the chip, you change the excitation for leakage current across the chip, it is the leakage current that causes a lot of the noise. at low ISO this is also chip voltage, and as a result, lower leakage current. It is the same reason that reducing the sensor area (i.e. each individual sensor) has increased noise, the impact of each electron of leakage current is much greater. Heat also makes this noise worse, although I have not seen, and am not suggesting people go out and bake or freeze their cameras to evaluate the impact of noise,
Loewell,

I've been studying various image sensor datasheets (Sony, Toshiba and others) and also various application notes over the last couple of hours. However, I cannot find any reference to what you said.

All designs show a seperate VGC (voltage gain control) curcuit, right after the ccd row readout circuits. To me that makes no difference. The amp sits in the sensor chip instead of in the processing chip, so what?

Do you have any further material to read?

- Bert
11-08-2007, 04:11 PM   #30
Veteran Member




Join Date: Jan 2007
Location: Northern California
Posts: 475
QuoteOriginally posted by bymy141 Quote
Loewell,

I've been studying various image sensor datasheets (Sony, Toshiba and others) and also various application notes over the last couple of hours. However, I cannot find any reference to what you said.

All designs show a seperate VGC (voltage gain control) curcuit, right after the ccd row readout circuits. To me that makes no difference. The amp sits in the sensor chip instead of in the processing chip, so what?

Do you have any further material to read?

- Bert
I think that CCD data is simply read out when you want to capture an image and that all further processing starts after that.

I also think that all such data is amplified at some point in the downstream signal path. It is possible to have a amp with a set gain designed for the lowest possible noise and then just multiply in the digital domain for gain, but I am pretty certain that the K10D uses a Programmable Gain Amplifier, which is a post-sensor analog amplifier that can be addressed to change the gain.

Perhaps there is more on-chip processing with CMOS sensors, which might include amplification, but as you say, it really does not matter where you do it, it is after you have read the data out of the pixel, or post-pixel, if you will.

Ray
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, dslr, iso, k10d, photography
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
K10d Iso 1600 tarsus Post Your Photos! 5 05-25-2008 09:39 PM
ISO 1600 Performance dylansalt Pentax DSLR Discussion 14 03-24-2008 08:34 PM
K10D at ISO 1600 Canada_Rockies Pentax DSLR Discussion 21 01-02-2008 06:04 AM
K10D at iso 1600 Kemal Pentax DSLR Discussion 15 10-21-2007 12:37 PM
1600 ISO K10D new picture ercole69 Pentax DSLR Discussion 0 11-07-2006 04:26 AM



All times are GMT -7. The time now is 12:34 AM. | See also: NikonForums.com, CanonForums.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