Ugly Hedgehog - Photography Forum
Home Active Topics Newest Pictures Search Login Register
Main Photography Discussion
Further on ISO invariance
Page 1 of 4 next> last>>
May 28, 2016 13:19:23   #
TriX Loc: Raleigh, NC
 
I've re-read Ron's original thread on this subject a number of times, and while I get the concept, I'm still struggling with one issue, and it appears to depend on where the ISO or amplifier-chain gain is controlled and the effect on the effective dynamic range of the sensor. I'd like to understand this at the chip/circuit level rather than by example shots. For background, I've read the spec sheets on a number of sensors ( but can't yet locate the specific sensor for the Nikon cameras referenced ) - both cmos and ccd. Some have the ADCs on-board the imager, while others are out-board. Some allow you to control the gain "pre ADC", some allow you to set the ADC comparator voltage (as was mentioned in the last post), and some control the post-ADC amplifier gain to control sensitivity and ISO. It would appear to me that depending on where the gain is controlled (pre, at or post ADC) that by under-exposing, you may limit the usable dynamic range of the output to something less than the available range of the sensor. What are your thoughts on this? Thanks in advance for any assistance in understanding.

Reply
May 28, 2016 15:17:27   #
selmslie Loc: Fernandina Beach, FL, USA
 
In the original thread BebuLamar wrote:
Regardless of how the amplification is done, doing so will limit the dynamic range of the sensor. It's simple when you double the ISO to twice the base ISO you only digitize the half of the dynamic range of the sensor. Instead of represent the entire dynamic range with 14 bit quantization it only use half the dynamic range. However doing so ensure that you have enough bits for smooth tonal gradation. If you didn't boost the ISO you can simply drop the most significant bit in the RAW files and have a 13 bit with the same result.
Regardless of how the amplification is done, doing... (show quote)

It seems that several posters to Rongnongno's thread, including the OP, fail to grasp the fundamental fallacy in ISO invariance.

Underexposing by 3 stops at base ISO (100?) takes advantage of only 1/8th of the sensor's physical capacity and in effect turns a 14-bit sensor into an 11-bit sensor. It uses the same exposure and generates the same S/N ratio as if you had exposed normally at ISO 800 in place of 100 and used all 14 bits.

You can recover the shadow information by applying a +3 exposure adjustment to make the image look the same as if you had captured it at ISO 800. However that only works if you image would have been relatively free of noise at ISO 800 in the first place.

The image must also not have any large areas of smooth tonal gradation in the shadows because underexposing and amplifying later will increase the chances of banding (posterization) there because there were not enough distinct levels in the lower bits. Banding is something we might see in carelessly edited 8-bit files in blue skies and especially after a B&W conversion.

The main advantage of underexposing is that you reduce the chance of blown highlights, assuming that you care.

The disadvantage is that you cannot review the image on the spot (the same is true for ETTR) so you might be disappointed when you start to convert the raw file.

Reply
May 28, 2016 15:21:04   #
Rongnongno Loc: FL
 
selmslie wrote:
.../...

And the ... strikes again. Nevermind the posts made using original files.

Reply
 
 
May 28, 2016 15:34:14   #
selmslie Loc: Fernandina Beach, FL, USA
 
Rongnongno wrote:
And the ... strikes again. Nevermind the posts made using original files.

I saw the posted examples and they demonstrate exactly what I stated.

You can recover shadows from a severely underexposed image but you risk also recovering noise and having poor definition of smooth dark tonality.

Examples of ETTR also show recovery of highlights, so long as they are not hopelessly blown.

The problem with the original thread is that it is based on a lack of understanding of the fundamentals of exposure compensation during the capture and later during raw development.

Until you are familiar with the trade-off, you would do well to stick to exposing normally and using +/-1 stop of EC for snow and coal piles.

Reply
May 29, 2016 03:57:39   #
Apaflo Loc: Anchorage, Alaska
 
selmslie wrote:
The main advantage of underexposing is that you reduce the chance of blown highlights, assuming that you care.

The disadvantage is that you cannot review the image on the spot (the same is true for ETTR) so you might be disappointed when you start to convert the raw file.

There is no technical advantage to underexposing. None.

Most people don't seem to understand precisely what "ISO invariance" is, and there are all sorts of claims about which cameras are or are not. Here is what Thom Hogan said:

"An ISO invariant sensor is going to have a dynamic range chart
that has a constant slope downwards as you increase ISO values.
An ISO variant camera is going to have ups and downs ... "
-- http://www.dslrbodies.com/cameras/the-d5d500-blog/iso-variance.html

The key words are "constant slope". That means the curve will be essentially a straight line for any portion which can be called "ISO invariant".

Below is an image showing charts for two cameras, one of which is perhaps the least "invariant" DSLR that I know of, and one of which is fairly near to a straight line. (I've trimmed the screen shot to avoid identifying the exact models. Note that the graph is from Bill Claff's webpage.)

On the webpage it is easy to get exact numbers for the sample points. Here is a chart showing values over a range from ISO 400 to ISO 800 for each camera,

ISO Top Graph Bottom Graph
400 9.63 8.03
503 9.33 7.68
636 8.97 6.43
800 8.64 7.30

For the top graph each higher step in ISO causes a linear change in the Dynamic Range. With the bottom graph, for the non ISO variant camera, the ISO change is not linear and the worst case is a 1/3 step higher from ISO 503 to ISO 636 actually reduces the Dynamic Range lower than it will be at ISO 800. Note that on even multiples such as 100, 200, 400, 800 and so on the change is perhaps not exactly 1 stop less Dynamic Range but fairly close to it (if we plotted only those values it would be a constant slope).

Okay, that is what defines ISO Invariance. Now the question is what significance is there? The answer is virtually none, and certainly not the claims that it means one can shoot at base ISO and get exactly the same image as if it were shot at a higher ISO!

Most of the comments to the effect that there is some advantage all center on shooting at something less than ISO 800. Note that the Dynamic Range, even with the lesser of these two cameras, is 7.3 fstops at that ISO. Most JPEG images have less than 7 fstops of Dynamic Range and most prints less than 5. That means unless significant gamma curve manipulation is done there will not be any visible noise in the shadows (the black point will be higher than the read noise).

The supposed equivalence is an artifact of the image not being displayed with enough Dynamic Range to see the noise and has no relevance to "ISO Invariance". The effect would be essentially identical with both of these cameras!

The advantages of using analog gain before the ADC would be seen for both cameras. The disadvantages of digital multiplication in correcting an underexposed image would also be identical for both cameras.

The entire concept that ISO non-variance is of some value is pseudo science based on attributing sometimes valid (or not) observations to the wrong cause.



Reply
May 29, 2016 07:25:15   #
selmslie Loc: Fernandina Beach, FL, USA
 
Apaflo wrote:
There is no technical advantage to underexposing. None. ...

Nobody has claimed that there is one. The point that has been clearly demonstrated is that there is little or no disadvantage visible in the image.

It is extremely difficult to show any any decline in shadow tonality as a result of having fewer steps in the raw file's in the tonal range. And while there is the potential for additional noise, when you start from base ISO and underexpose by three stops with a good sensor, you still do not reach S/N levels that are visible or not easily repaired.

The advantage that is claimed is only one of convenience.
Apaflo wrote:
... Most people don't seem to understand precisely what "ISO invariance" is, and there are all sorts of claims about which cameras are or are not. ...

That's true since it is a fuzzy concept that has little practical relevance. Most people do not care what it is. They simply don't care what you, Thom Hogan or anyone else says about it. They look at images, not graphs.

What really upsets people that have been drinking the ETTR Kool Aid for so many years is that, within limits, underexposing can be just as easily adjusted in post processing as overexposing, without dire consequences becoming visible in the image.

Since that has been clearly demonstrated, post processing to correct exposure errors is a viable first step and there is no need to obsess over ETTR or ETTL methodologies. The more reasonable approach would be to try to get the exposure right so that information can easily be recovered from both the shadows and the highlights.

Reply
May 29, 2016 07:56:22   #
oldtigger Loc: Roanoke Virginia-USA
 
selmslie wrote:
try to get the exposure right so that information can easily be recovered from both the shadows and the highlights.

Sounds like a reasonable solution to a common problem.

Reply
 
 
May 29, 2016 08:30:33   #
BebuLamar
 
oldtigger wrote:
Sounds like a reasonable solution to a common problem.


Is there a problem? I don't see it!

Reply
May 29, 2016 08:50:46   #
Apaflo Loc: Anchorage, Alaska
 
BebuLamar wrote:
Is there a problem? I don't see it!

When the advice is to set ISO to 100 and not worry about under exposure... there is a big problem!

Reply
May 29, 2016 08:55:21   #
BebuLamar
 
Apaflo wrote:
When the advice is to set ISO to 100 and not worry about under exposure... there is a big problem!


Doing so would cause underexposure but one has to follow the advice.

Reply
May 29, 2016 09:14:38   #
kymarto Loc: Portland OR and Milan Italy
 
For me the issue is quite clear. If one has time to very accurately assess the brightest highlight in the frame that one doesn't want blown out, and knowing exactly how much headroom one has before that highlight becomes unrecoverable or exhibits false color, then it makes perfect sense to expose for that highlight.

That works perfectly for a shot of a static subject in unchanging light, when one has the time to accurately identify and meter that highlight. In real life, I come across that particular situation very rarely. In the real world, outside, the light is constantly changing or the subject is moving or both, and I often have less than a second to frame and shoot when action is happening.

With my Nikon D800E, I have done tests and found little visible difference shooting at the base ISO of 100 and underexposing two to three stops from what the matrix meter is telling me, and then adjusting in post, as compared to shooting at the bleeding edge of the brightest highlight, which is something like 1.5 stops over the usual metered exposure. Yes, if I enlarge to 100% and look closely, there is a minor difference, but I would rather suffer that almost-invisible degradation in 100 shots than lose one good shot to blown highlights. Even with that compromise, my shadows are cleaner than those of many--if not most--of the cameras on the market. For me it's a complete no-brainer.

Here are a couple of studies of small waves in the late afternoon. These were "underexposed" (according to the automatic matrix metering) by 3 EV. I was easily able to recover shadow detail, and not blow out the highlights, or at least limit the blowout to acceptable (to me) levels:


(Download)


(Download)


(Download)


(Download)

Reply
 
 
May 29, 2016 09:53:24   #
selmslie Loc: Fernandina Beach, FL, USA
 
Apaflo wrote:
When the advice is to set ISO to 100 and not worry about under exposure... there is a big problem!

A big problem? Really?

Here are five images taken at ISO 100 on an A7 II. I used EC on the camera and offset it in Capture One Pro 9. At EC -3 the red umbrella started to lose saturation and at EC +3 the highlights started to get blown.

Incidentally, I got blinkies on EC-2, EC+1 and EC+2 and none on EC-1 or EC+0. On the camera, the blue JPEG histogram only reached the right edge for EC+2.

You might need to pixel peep to see any qualitative difference among these three images.

So where is the big problem?

EC -2
EC -2...
(Download)

EC -1
EC -1...
(Download)

No EC on camera
No EC on camera...
(Download)

EC +1
EC +1...
(Download)

EC +2
EC +2...
(Download)

Reply
May 29, 2016 10:23:48   #
zigipha Loc: north nj
 
To answer the OPs question
1. photons enter a photosite. The more photons enter, the higher the voltage.
2. At the end of the exposure time, the voltage (analog) for each photosite is read, passing through an analog amplifier, and presented to an analog to digital converter
3. The analog amplifier gives the appearance of increasing the sensitivity of the sensor (in fact it does not, but the amplifier kicks in when you change the iso setting).
4. Ideally, the amplifier should be on the sensor "pre ADC" in your terms.
5. Alternatively, the amplifier can be on the ADC itself. An alternate way to simulate analog amplification is to change the reference voltage on the ADC. These two approaches give basically the same result.
Recap - on-sensor amp, or pre-ADC amp, or change ADC reference voltage, are basically the same (they are not exactly the same but close enough compared to digital amplification)
6. The post ADC amplifier works on the digital data. This means that the ADC was not being used to its maximum dynamic range and therefore you have worse quantization noise than the analog method of amplifying the sensor voltage.
Recap - post ADC amplification is worse than pre-ADC amplification.
Recap2 - post ADC amplification is the same as you changing the brightness level in post processing.

Reply
May 29, 2016 11:01:04   #
selmslie Loc: Fernandina Beach, FL, USA
 
selmslie wrote:
... At EC -3 the red umbrella started to lose saturation and at EC +3 the highlights started to get blown. ...

For the sake of completeness, here are the two other images.

EC -3 on the camera, +2.67 in post processing
EC -3 on the camera, +2.67 in post processing...
(Download)

EC+3 on the camera, -3 in post processing
EC+3 on the camera, -3 in post processing...
(Download)

Reply
May 29, 2016 11:06:31   #
TriX Loc: Raleigh, NC
 
zigipha wrote:
To answer the OPs question
1. photons enter a photosite. The more photons enter, the higher the voltage.
2. At the end of the exposure time, the voltage (analog) for each photosite is read, passing through an analog amplifier, and presented to an analog to digital converter
3. The analog amplifier gives the appearance of increasing the sensitivity of the sensor (in fact it does not, but the amplifier kicks in when you change the iso setting).
4. Ideally, the amplifier should be on the sensor "pre ADC" in your terms.
5. Alternatively, the amplifier can be on the ADC itself. An alternate way to simulate analog amplification is to change the reference voltage on the ADC. These two approaches give basically the same result.
Recap - on-sensor amp, or pre-ADC amp, or change ADC reference voltage, are basically the same (they are not exactly the same but close enough compared to digital amplification)
6. The post ADC amplifier works on the digital data. This means that the ADC was not being used to its maximum dynamic range and therefore you have worse quantization noise than the analog method of amplifying the sensor voltage.
Recap - post ADC amplification is worse than pre-ADC amplification.
Recap2 - post ADC amplification is the same as you changing the brightness level in post processing.
To answer the OPs question br 1. photons enter a p... (show quote)


Thank you - those were exactly my thoughts as well. I designed, built and sold digitizers for many years, but it's hard to find camera schematics (or even block diagrams) so I can see exactly what's going on internally regarding changing the ISO setting.

Reply
Page 1 of 4 next> last>>
If you want to reply, then register here. Registration is free and your account is created instantly, so you can post right away.
Main Photography Discussion
UglyHedgehog.com - Forum
Copyright 2011-2024 Ugly Hedgehog, Inc.