Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
02-24-2019, 08:37 PM - 2 Likes   #16
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
QuoteOriginally posted by photoptimist Quote
You "gain" potentially ugly banding in flat areas such as the sky.

Imagine you under-expose an image by using an ISO that is 5-stops low and then multiply by 32 (= 5 stops) in PP to boost it. Now, instead of adjacent pixels in smooth areas differing by an unnoticeable 1 gray value, they differ by a jump of least 32 gray values.

This is why I'm skeptical that cameras use "post ADC multiplication" to any great extent -- it would create both artifacts in the image and a very strange spiky histogram.
I have utilised Iso invariance by staying at base Iso for quite a while now and often have to push 4 or 5 stops in PP. I have never seen any sign of banding or for that matter any indication of it in the histogram. I occasionally used to get a little banding when I was using 8bit Gimp as an editor so I know what I am looking for. The main downside of staying at base Iso is it takes fairly precise control of the histogram in the raw developer to get the best out of it. But then the control you have of the highlights makes it all worthwhile.

02-25-2019, 03:54 PM   #17
chd
Forum Member
chd's Avatar

Join Date: Dec 2018
Location: Tucson AZ
Posts: 54
Original Poster
QuoteOriginally posted by GUB Quote
I have utilised Iso invariance by staying at base Iso for quite a while now and often have to push 4 or 5 stops in PP.
Just to make sure I understand -- this means leaving the camera ISO set at 100, or whatever its lowest value is, using the aperture and shutter speed values you want for whatever other reasons, and allowing the result to be underexposed possibly by several stops (at least according to the camera's judgment)? And then bringing it back up in Photoshop afterward, as you say.

If I'm interpreting that correctly, it seems that this would have worse results, at least to some extent -- by skipping or minimizing an early analog gain amplification, you are often feeding a pretty low-level signal to the analog-digital-converter. Since you aren't using the full input range of the ADC, you are getting more quantization errors/noise and less resolution in the conversion than you would otherwise, no?
02-25-2019, 04:37 PM   #18
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
QuoteOriginally posted by chd Quote
Just to make sure I understand -- this means leaving the camera ISO set at 100, or whatever its lowest value is, using the aperture and shutter speed values you want for whatever other reasons, and allowing the result to be underexposed possibly by several stops (at least according to the camera's judgment)? And then bringing it back up in Photoshop afterward, as you say.

If I'm interpreting that correctly, it seems that this would have worse results, at least to some extent -- by skipping or minimizing an early analog gain amplification, you are often feeding a pretty low-level signal to the analog-digital-converter. Since you aren't using the full input range of the ADC, you are getting more quantization errors/noise and less resolution in the conversion than you would otherwise, no?
My understanding is on a basic level the process goes from the sensor to analogue gain (Iso) then converted to digital signal which is what you PP on. On a non Iso invariant sensor there is noise input from the anloque/digital conversion that PPing amplifies. In an Iso invariant sensor there is negligible noise input from the conversion which means PPing has to all intents and purposes the same output as Iso. In practise you always end up with a difference between the two processes but neither is necessarily a better IQ image. But of course by using the curve tool in the histogram you can have far more control of dynamic range in Iso 100 images.
EDIT check the link that Adam supplied and you will see a good example of a comparison between the two that I did. (post 18)
02-25-2019, 04:54 PM   #19
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
QuoteOriginally posted by chd Quote
Just to make sure I understand -- this means leaving the camera ISO set at 100, or whatever its lowest value is, using the aperture and shutter speed values you want for whatever other reasons, and allowing the result to be underexposed possibly by several stops (at least according to the camera's judgment)? And then bringing it back up in Photoshop afterward, as you say.
Exactly right. It creates good process in that you start concentrating on the optimum aperture and shutter setting for the situation. So basically if you have the aperture as open as appropriately possible and the shutter as slow as appropriately possible and you have checked that your highlights are not clipped then you are at best practise. How simple is that.
Rather than staying at base Iso you can underexpose by say 3 stops via Iso and end up with a visible image to chimp on and easier pping and still keep most of that control of your dynamic range.

02-25-2019, 09:26 PM   #20
chd
Forum Member
chd's Avatar

Join Date: Dec 2018
Location: Tucson AZ
Posts: 54
Original Poster
QuoteOriginally posted by GUB Quote
Exactly right. It creates good process in that you start concentrating on the optimum aperture and shutter setting for the situation. So basically if you have the aperture as open as appropriately possible and the shutter as slow as appropriately possible and you have checked that your highlights are not clipped then you are at best practise. How simple is that.
Rather than staying at base Iso you can underexpose by say 3 stops via Iso and end up with a visible image to chimp on and easier pping and still keep most of that control of your dynamic range.
Ok... so, what about the concern that I mentioned before? You are sending a low-level signal to the ADC and therefore you are getting much less resolution in its digital output than you would if you sent it closer to a full-range signal... am I describing the concern clearly?
02-25-2019, 10:00 PM   #21
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
Yes you are clear but think about it. The amount of image data collected is at the sensor and achieved before the Iso amplification. Therefore the Iso amplified result going to the ADC contains no more info than the base level info. What Photoptemist seems to be implying is that the analogue amplification at Iso time is somehow superior to the digital amplification at pp time. I haven't heard that argument before nor can I see it in results.
02-26-2019, 12:04 PM   #22
chd
Forum Member
chd's Avatar

Join Date: Dec 2018
Location: Tucson AZ
Posts: 54
Original Poster
QuoteOriginally posted by GUB Quote
Yes you are clear but think about it. The amount of image data collected is at the sensor and achieved before the Iso amplification. Therefore the Iso amplified result going to the ADC contains no more info than the base level info. What Photoptemist seems to be implying is that the analogue amplification at Iso time is somehow superior to the digital amplification at pp time. I haven't heard that argument before nor can I see it in results.
It may not be generally visible in results; I accept that the question may be academic. That doesn't make it less interesting for me. :-)

You're right that the data coming off the sensor does not have more information in it after amplification than it does before, but for a low level signal (e.g. a very dark/"underexposed" image) that data is compressed into a much smaller range, and the variations are much closer together. Let's say---and I'm making up the numbers here, but just for illustration---let's say that the possible output range of the sensor is +- 5 volts. For an image where a few whites are blown and a few blacks are clipped and most of the rest is in between, when you read the data straight from the sensor you'll get a distribution of values along pretty much the whole +-5V range. Your ADC is going to take this input and fit it into the whole range of values available in its 16-bit (or whatever) output.


Now let's say you have the same scene, same settings, just much less light. One thing that happens is more blacks will be clipped, but you're correct that amplification won't fix that if the sensor never recorded it. However, another thing that happens is that the +-5V signal range is now all contained within, say, -5V to -4V. That's what you're feeding to an ADC designed to accept +-5V. Now the ADC is going to take the same range of signal you gave it in the first case, but it's going to fit it into the bottom 8 bits, say, because it's all very low level and it can't normalize it because it doesn't know that the next pixel isn't going to be white, and they all have to scale together. Lost resolution.

The analog amplification stage does matter because it more or less normalizes the input to the ADC. The amount of amplification has to be fixed across an image though, which is why you have to set it ahead of time and it can't be done on the fly.

TL,DR; --- the analog amplification doesn't get any more resolution from the sensor, but it does get more resolution from the analog to digital conversion.

02-26-2019, 01:41 PM   #23
Loyal Site Supporter
Loyal Site Supporter




Join Date: Jun 2009
Location: Tumbleweed, Arizona
Photos: Gallery | Albums
Posts: 5,699
.... and here is another take on the topic. He mostly agrees, but takes issue with the variant sensors and invariant sensors.


02-27-2019, 02:23 AM   #24
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
QuoteOriginally posted by interested_observer Quote
.... and here is another take on the topic. He mostly agrees, but takes issue with the variant sensors and invariant sensors.


I agree fully with that video with the one exception that I thought that the D850 with its non Sony sensor was not Iso invariant.
02-27-2019, 03:22 AM   #25
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
QuoteOriginally posted by chd Quote
One thing that happens is more blacks will be clipped
The thing is that Black end doesn't seem to clip like the light end. The info seems to just compress up against that end waiting to be drawn out.
OK -- you have an Iso invariant sensor
By definition this means no noise is introduced nor information lost between Amp Iso and Amp PP.
If we accept both Amps are equal in efficiency
Then why can it matter which Amp we use.

If you go online and research you will find general acknowledgement of the Iso invariance principle. Even the Canon people don't dispute it's existence despite their cameras being Iso variant. They just dispute its importance. Here - take a look at a site that loves to dis Pentax and look at their Iso invariance section. Just ignore their dribblespeak about the SonyA7ii.
Pentax K-1 Review: Digital Photography Review
02-27-2019, 09:50 AM   #26
chd
Forum Member
chd's Avatar

Join Date: Dec 2018
Location: Tucson AZ
Posts: 54
Original Poster
QuoteOriginally posted by GUB Quote
By definition this means no noise is introduced nor information lost between Amp Iso and Amp PP.
I'm not sure exactly what you're referring to here by "Amp ISO" and "Amp PP". Can you be more specific? Is "Amp ISO" the analog amplification stage between the sensor and the ADC, maybe? What's "Amp PP" - is that the ADC? If not, what are you referring to?

It's impossible not to introduce any noise in these stages, though a good design tries to minimize it. There is always read noise.
02-27-2019, 12:45 PM   #27
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
QuoteOriginally posted by chd Quote
I'm not sure exactly what you're referring to here by "Amp ISO" and "Amp PP". Can you be more specific? Is "Amp ISO" the analog amplification stage between the sensor and the ADC, maybe? What's "Amp PP" - is that the ADC? If not, what are you referring to?

It's impossible not to introduce any noise in these stages, though a good design tries to minimize it. There is always read noise.
Amp Iso and ApPP Applying gain at the Iso point in the chain and applying gain at the Post processing point. (PP).
The definition of an Iso invariant sensor revolves around the requirement that there is no significant noise introduced between AmpIso and AmpPP.
It is accepted that such a setup exists in the current Sony sensors amongst others.
02-27-2019, 02:20 PM   #28
chd
Forum Member
chd's Avatar

Join Date: Dec 2018
Location: Tucson AZ
Posts: 54
Original Poster
QuoteOriginally posted by GUB Quote
Amp Iso and ApPP Applying gain at the Iso point in the chain and applying gain at the Post processing point. (PP).
The definition of an Iso invariant sensor revolves around the requirement that there is no significant noise introduced between AmpIso and AmpPP.
It is accepted that such a setup exists in the current Sony sensors amongst others.
Got it, thanks.

I found an interesting and quite a bit more rigorous (but still entirely accessible) exploration of this, and I think I now have my head around it quite a bit better. I definitely recommend this article:


Noise, Dynamic Range and Bit Depth in Digital SLRs

A few years old now, but the theory seems still entirely solid. As usual, there's a lot in the details that gets glossed over in many discussions, and "it's not quite that simple."
03-23-2019, 08:02 AM   #29
New Member




Join Date: Mar 2015
Posts: 17
Well just wanted to give my butter on the subject.

Lets look at what the different steps are and why ISO on digital cameras do make sense and that some types do not. Here is a simplified way of object to data. And a quick overview of the factors that thereof influence the image quality in 3 Steps:
Object ---> Medium(usually air) ------------------------------> Lens -------------------> filter on sensor for IR/UV ---> micro lenses --->Bayer Filter-------------------------------------------> Sensor --------->
-------------|Dust, Humidity, Temperature differences… ---| Optical errors, etc. ---| Filters out unwanted light ----|------------------- | Lowers luminosity and sharpness for colour--- |



---> Amplifier on each pixel(Mostly CMOS) --------------------------> Amplifier after processor -------------------------- ----------------------------------> Analogue processing and conversion to digital signal converter--------->
-----|First Factor for A-ISO, k-1 ISO happens here I assume ----- | This is the K-1 mark II ISO improvements happen I assume -----------------| Creates NOISE!!!!!

|||||| This ISO Amplification makes sense since it is before new Noise gets generated |||||

-------------------------------------------------------------------------------------------------------------


--> Real RAW data ---> Magic Processing -------------------------------------------------------------------------------------------------- ----------------> RAW FILE -------------> Conversion to JPEG
-----------------------------| Here a lot of magic proprietary stuff and the ISO multiplication that can be done theoretically in post happens--|---------------------------- | Loss in DATA over and over in each save.


So why is the PENTAX almost ISO invariant, well digital cameras have improved a lot in these years and the Noise generated post sensor, has become relatively low. Also programs like photoshop apply noise filters in background that hides noise, that is why you will get similar results results shooting at iso 100. But ISO Analogue amplification still make a lot of sense as it avoids a lot of later noise.


I from experience in experiments outside of fotography know how important such early amplification of low signals can help with a lot better cleaner results.

Please also look at this article:
Is ISO Fake? | DSLRBodies | Thom Hogan

Last edited by elicius; 03-23-2019 at 08:11 AM.
03-29-2019, 01:48 AM   #30
GUB
Loyal Site Supporter
Loyal Site Supporter
GUB's Avatar

Join Date: Aug 2012
Location: Wanganui
Photos: Gallery | Albums
Posts: 5,735
The very definition of an Iso invariant camera is :

Analogue processing and conversion to digital signal converter--------->
Creates NO NOISE!!!!!

Iso invariance is an accepted attribute of the later Sony sensors (among others)
It is not hard to go and test the principle yourself. Just take a correctly exposed shot of a setup that requires a high Iso like 3200 and take a second shot with aperture and shutter unchanged but at Iso 100. Push the Iso 100 shot 5 stops in your raw developer and compare the results.
People should stop debating whether it exists or not but rather discuss whether it can be an important part of their processing.
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!
24x36mm, analog, cameras, full-frame, iso, k-1, pentax, settings
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
A Close Look at the Pentax K-1's Internals PF Staff Homepage & Official Pentax News 1 02-29-2016 04:31 AM
What is this piece called? From Pentax lens internals Harrison Pentax SLR Lens Discussion 18 11-13-2013 08:08 AM
Info: How the internals of K-01 look like raider Pentax K-01 2 08-18-2012 04:24 AM
who knows Pentax 35mm internals?-rangefinder jaydag71 Pentax Camera and Field Accessories 1 08-20-2007 11:31 AM



All times are GMT -7. The time now is 07:21 PM. | 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