DXOMark ISO sensitivity - confused...

Started Jun 5, 2021 | Questions
spider-mario
spider-mario Senior Member • Posts: 1,000
Re: History and Evolution of ISO 12232
1

57even wrote:

spider-mario wrote:

57even wrote:

My understanding was that the RGB 118 actually applied to the Y channel of the YCbCr version of the JPEG file. So is it chroma independent and would apply to any RGB space.

The way I read it, the Y that they specify is unrelated the Y′ of Y′CbCr. It’s just their specification of how to combine an RGB triplet into a single value for the purpose of measurement. It assumes sRGB primaries and transfer function but has no effect if R=G=B anyway (Y will just be equal to them).

I am assuming that you are referring to this section, please correct me otherwise:

Isn't that the standard encoding for YCbCr luminance for sRGB?

If not, then I guess this article is wrong...

http://dougkerr.net/Pumpkin/articles/SOS_REI.pdf

Y′CbCr doesn’t have the intermediary linearization that CIPA’s document has. If the image is in sRGB, then CIPA’s “Y” will be the “correct” luminance but with sRGB’s non-linearity applied, whereas Y′ may not be.
For example, if (R, G, B) = (0, 128/255, 0):

  • CIPA linearizes the G from 128/255 to 55/255, multiplies it by 0.7152 to obtain the correct luminance of ~0.15, and then goes back to sRGB non-linearity, giving approximately 110/255 if my calculations are correct.
  • Y′CbCr applies 0.7152 to 128/255 directly, giving ~92/255.
 spider-mario's gear list:spider-mario's gear list
Canon G1 X III Olympus TG-6 Olympus E-M1 II Olympus E-M1 III Canon EOS R6 +14 more
Jack Hogan Veteran Member • Posts: 8,064
Re: DXOMark ISO sensitivity - confused...

57even wrote:

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote: ...

I think another way to look at it is that 78 comes from pi/4*100 because of the interaction between the metering and luminance formulas.

OK, you mean the ratio of image plane illuminance to luminance?

Hm = mean exposure = pi/4*Lm*t/N^2

with Lm mean luminance, N f-number and t exposure time. A reflected light meter will produce exposure settings based on

N^2/t = Lm*S/k

with S ISO speed and k the meter calibration constant. Combining the two:

S = pi/4*k/Hm

Hm is 10/S for digital cameras according to ISO 2240 and 2721 so k is 40/pi = 12.73 (Minolta, CaNikon light meters etc.). Choose your poison as far as saturation exposure Hsat is concerned, say with a lot of hand waving Hm = 0.128 Hsat. Then

S = pi/4*12.73/.128/Hsat = 78.1/Hsat.

There is some recursion there but I trust you get the idea.

Jack

Would that be the same for sensor plane illuminance (with a different constant)?

Assuming my 'meter' is the sensor in this case.

Yes, though the choice of constant is a bit more arbitrary for incident light meters.  Camera meters are of the reflected light type so Luminance tends to be more relevant for this type of analysis.

OP (unknown member) Forum Pro • Posts: 16,732
Re: DXOMark ISO sensitivity - confused...

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote: ...

I think another way to look at it is that 78 comes from pi/4*100 because of the interaction between the metering and luminance formulas.

OK, you mean the ratio of image plane illuminance to luminance?

Hm = mean exposure = pi/4*Lm*t/N^2

with Lm mean luminance, N f-number and t exposure time. A reflected light meter will produce exposure settings based on

N^2/t = Lm*S/k

with S ISO speed and k the meter calibration constant. Combining the two:

S = pi/4*k/Hm

Hm is 10/S for digital cameras according to ISO 2240 and 2721 so k is 40/pi = 12.73 (Minolta, CaNikon light meters etc.). Choose your poison as far as saturation exposure Hsat is concerned, say with a lot of hand waving Hm = 0.128 Hsat. Then

S = pi/4*12.73/.128/Hsat = 78.1/Hsat.

There is some recursion there but I trust you get the idea.

Jack

Would that be the same for sensor plane illuminance (with a different constant)?

Assuming my 'meter' is the sensor in this case.

Yes, though the choice of constant is a bit more arbitrary for incident light meters. Camera meters are of the reflected light type so Luminance tends to be more relevant for this type of analysis.

How does that work with mirrorless cameras?

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

OP (unknown member) Forum Pro • Posts: 16,732
Re: History and Evolution of ISO 12232

spider-mario wrote:

57even wrote:

spider-mario wrote:

57even wrote:

My understanding was that the RGB 118 actually applied to the Y channel of the YCbCr version of the JPEG file. So is it chroma independent and would apply to any RGB space.

The way I read it, the Y that they specify is unrelated the Y′ of Y′CbCr. It’s just their specification of how to combine an RGB triplet into a single value for the purpose of measurement. It assumes sRGB primaries and transfer function but has no effect if R=G=B anyway (Y will just be equal to them).

I am assuming that you are referring to this section, please correct me otherwise:

Isn't that the standard encoding for YCbCr luminance for sRGB?

If not, then I guess this article is wrong...

http://dougkerr.net/Pumpkin/articles/SOS_REI.pdf

Y′CbCr doesn’t have the intermediary linearization that CIPA’s document has. If the image is in sRGB, then CIPA’s “Y” will be the “correct” luminance but with sRGB’s non-linearity applied, whereas Y′ may not be.
For example, if (R, G, B) = (0, 128/255, 0):

  • CIPA linearizes the G from 128/255 to 55/255, multiplies it by 0.7152 to obtain the correct luminance of ~0.15, and then goes back to sRGB non-linearity, giving approximately 110/255 if my calculations are correct.
  • Y′CbCr applies 0.7152 to 128/255 directly, giving ~92/255.

I was referring to YCbCr, which I assumed was non-linear, as opposed to Y'CbCr.

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

Jack Hogan Veteran Member • Posts: 8,064
Re: DXOMark ISO sensitivity - confused...

57even wrote:

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote: ...

I think another way to look at it is that 78 comes from pi/4*100 because of the interaction between the metering and luminance formulas.

OK, you mean the ratio of image plane illuminance to luminance?

Hm = mean exposure = pi/4*Lm*t/N^2

with Lm mean luminance, N f-number and t exposure time. A reflected light meter will produce exposure settings based on

N^2/t = Lm*S/k

with S ISO speed and k the meter calibration constant. Combining the two:

S = pi/4*k/Hm

Hm is 10/S for digital cameras according to ISO 2240 and 2721 so k is 40/pi = 12.73 (Minolta, CaNikon light meters etc.). Choose your poison as far as saturation exposure Hsat is concerned, say with a lot of hand waving Hm = 0.128 Hsat. Then

S = pi/4*12.73/.128/Hsat = 78.1/Hsat.

There is some recursion there but I trust you get the idea.

Jack

Would that be the same for sensor plane illuminance (with a different constant)?

Assuming my 'meter' is the sensor in this case.

Yes, though the choice of constant is a bit more arbitrary for incident light meters. Camera meters are of the reflected light type so Luminance tends to be more relevant for this type of analysis.

How does that work with mirrorless cameras?

Don't forget, Dxo applies this to raw data, which is supposedly linear with sensor plane luminance.

But the ISO standard is all about rendered images, so everyone can pretty well do whatever they like and call it REI. Bit of a joke to call it a standard really.

On the other hand it wouldn't seem too difficult to approximate a reflected light meter given raw rgb data. Though I don't know anybody is doing this, since even in 2021 no manufacturer is deigning themselves to provide a raw histogram.

Jack

spider-mario
spider-mario Senior Member • Posts: 1,000
Re: History and Evolution of ISO 12232

57even wrote:

I was referring to YCbCr, which I assumed was non-linear, as opposed to Y'CbCr.

I included the ′ to be more “correct” but as far as I know, they refer to the same thing and JPEG doesn’t do the intermediary linearization found in CIPA’s document. It’s just a direct matrix transform from non-linear sRGB.

 spider-mario's gear list:spider-mario's gear list
Canon G1 X III Olympus TG-6 Olympus E-M1 II Olympus E-M1 III Canon EOS R6 +14 more
alanr0 Senior Member • Posts: 2,523
Re: History and Evolution of ISO 12232

57even wrote:

spider-mario wrote:

57even wrote:

My understanding was that the RGB 118 actually applied to the Y channel of the YCbCr version of the JPEG file. So is it chroma independent and would apply to any RGB space.

The way I read it, the Y that they specify is unrelated the Y′ of Y′CbCr. It’s just their specification of how to combine an RGB triplet into a single value for the purpose of measurement. It assumes sRGB primaries and transfer function but has no effect if R=G=B anyway (Y will just be equal to them).

I am assuming that you are referring to this section, please correct me otherwise:

Isn't that the standard encoding for YCbCr luminance for sRGB?

It may be, but as you see, CIPA DC-004 describes extraction of a luminance component from RGB components, not use of the Y component of a YCbCr signal.

ISO 12232:2006 makes no reference to either EXIF or to a JPEG file format.

It states: "For a colour camera, the camera white balance should be adjusted, if possible, to provide proper white balance (equal RGB signal levels) for the illumination light source, as specified in ISO 14524."

Specifically for the noise-based measurement: "The noise of the luminance and colour difference signals shall be determined from CRT display output-referred RGB colour signals based on the ITU-R BT.709 RGB primaries and white point, such as the sRGB and sYCC signals defined in IEC 61966-2-1, which are used as output signals in many DSCs".

So extended gamut sYCC, a YCbCr transformation based on ITU-R BT.601 is an option, but not a requirement for the noise-based measurement.

More to the point, for the SOS measurement, there is no transformation specified. As spider-mario explains above, provided the white balance is adjusted in compliance with section 5.4, then R=G=B, and no transformation is needed.

If not, then I guess this article is wrong...

http://dougkerr.net/Pumpkin/articles/SOS_REI.pdf

In a footnote he says "Unfortunately, ISO 12232 neglects to mention this, making its definition of SOS meaningless. In the CIPA standard it is covered only in an annex."

I think he is reading his own interpretation into the standards. The CIPA standard explicitly states that the explanatory annex he refers to is not part of the standard.

So it is a requirement in neither standard, and is not needed if the ISO 12232 test conditions regarding white balance are complied with.

-- hide signature --

Alan Robinson

OP (unknown member) Forum Pro • Posts: 16,732
Re: History and Evolution of ISO 12232

alanr0 wrote:

57even wrote:

spider-mario wrote:

57even wrote:

My understanding was that the RGB 118 actually applied to the Y channel of the YCbCr version of the JPEG file. So is it chroma independent and would apply to any RGB space.

The way I read it, the Y that they specify is unrelated the Y′ of Y′CbCr. It’s just their specification of how to combine an RGB triplet into a single value for the purpose of measurement. It assumes sRGB primaries and transfer function but has no effect if R=G=B anyway (Y will just be equal to them).

I am assuming that you are referring to this section, please correct me otherwise:

Isn't that the standard encoding for YCbCr luminance for sRGB?

It may be, but as you see, CIPA DC-004 describes extraction of a luminance component from RGB components, not use of the Y component of a YCbCr signal.

ISO 12232:2006 makes no reference to either EXIF or to a JPEG file format.

It states: "For a colour camera, the camera white balance should be adjusted, if possible, to provide proper white balance (equal RGB signal levels) for the illumination light source, as specified in ISO 14524."

Specifically for the noise-based measurement: "The noise of the luminance and colour difference signals shall be determined from CRT display output-referred RGB colour signals based on the ITU-R BT.709 RGB primaries and white point, such as the sRGB and sYCC signals defined in IEC 61966-2-1, which are used as output signals in many DSCs".

So extended gamut sYCC, a YCbCr transformation based on ITU-R BT.601 is an option, but not a requirement for the noise-based measurement.

More to the point, for the SOS measurement, there is no transformation specified. As spider-mario explains above, provided the white balance is adjusted in compliance with section 5.4, then R=G=B, and no transformation is needed.

If not, then I guess this article is wrong...

http://dougkerr.net/Pumpkin/articles/SOS_REI.pdf

In a footnote he says "Unfortunately, ISO 12232 neglects to mention this, making its definition of SOS meaningless. In the CIPA standard it is covered only in an annex."

I think he is reading his own interpretation into the standards. The CIPA standard explicitly states that the explanatory annex he refers to is not part of the standard.

So it is a requirement in neither standard, and is not needed if the ISO 12232 test conditions regarding white balance are complied with.

Gotcha.

Interestingly, his article is reference by the Wiki article on ISO.

Easy misunderstanding though, given that they call luminance 'Y' rather than just luminance.

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

OP (unknown member) Forum Pro • Posts: 16,732
Re: History and Evolution of ISO 12232

spider-mario wrote:

57even wrote:

I was referring to YCbCr, which I assumed was non-linear, as opposed to Y'CbCr.

I included the ′ to be more “correct” but as far as I know, they refer to the same thing and JPEG doesn’t do the intermediary linearization found in CIPA’s document. It’s just a direct matrix transform from non-linear sRGB.

Yeah, I think Alan agrees with you

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

OP (unknown member) Forum Pro • Posts: 16,732
Re: DXOMark ISO sensitivity - confused...
1

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote:

57even wrote:

Jack Hogan wrote: ...

I think another way to look at it is that 78 comes from pi/4*100 because of the interaction between the metering and luminance formulas.

OK, you mean the ratio of image plane illuminance to luminance?

Hm = mean exposure = pi/4*Lm*t/N^2

with Lm mean luminance, N f-number and t exposure time. A reflected light meter will produce exposure settings based on

N^2/t = Lm*S/k

with S ISO speed and k the meter calibration constant. Combining the two:

S = pi/4*k/Hm

Hm is 10/S for digital cameras according to ISO 2240 and 2721 so k is 40/pi = 12.73 (Minolta, CaNikon light meters etc.). Choose your poison as far as saturation exposure Hsat is concerned, say with a lot of hand waving Hm = 0.128 Hsat. Then

S = pi/4*12.73/.128/Hsat = 78.1/Hsat.

There is some recursion there but I trust you get the idea.

Jack

Would that be the same for sensor plane illuminance (with a different constant)?

Assuming my 'meter' is the sensor in this case.

Yes, though the choice of constant is a bit more arbitrary for incident light meters. Camera meters are of the reflected light type so Luminance tends to be more relevant for this type of analysis.

How does that work with mirrorless cameras?

Don't forget, Dxo applies this to raw data, which is supposedly linear with sensor plane luminance.

But the ISO standard is all about rendered images, so everyone can pretty well do whatever they like and call it REI. Bit of a joke to call it a standard really.

Yup. But this is what happens when manufacturers write their own standards and then ask standards organisations to publish them

On the other hand it wouldn't seem too difficult to approximate a reflected light meter given raw rgb data. Though I don't know anybody is doing this, since even in 2021 no manufacturer is deigning themselves to provide a raw histogram.

Jack

I had assumed they approximated sensor plane illuminance from the video feed (handily converted to sRGB already).

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

alanr0 Senior Member • Posts: 2,523
Re: History and Evolution of ISO 12232
1

57even wrote:

alanr0 wrote:

57even wrote:

If not, then I guess this article is wrong...

http://dougkerr.net/Pumpkin/articles/SOS_REI.pdf

In a footnote he says "Unfortunately, ISO 12232 neglects to mention this, making its definition of SOS meaningless. In the CIPA standard it is covered only in an annex."

I think he is reading his own interpretation into the standards. The CIPA standard explicitly states that the explanatory annex he refers to is not part of the standard.

So it is a requirement in neither standard, and is not needed if the ISO 12232 test conditions regarding white balance are complied with.

Gotcha.

Interestingly, his article is reference by the Wiki article on ISO.

The Doug Kerr reference by Wikipedia is to the half-stop difference between saturation-based speed and Standard Output Sensitivity.

Easy misunderstanding though, given that they call luminance 'Y' rather than just luminance.

Wikipedia: Luminance: "Not to be confused with Luma (video), Luminescence, or Illuminance"

The ISO standard already uses "Luminance" in the context of "Scene Luminance".  If they needed to refer to the Y component, I would expect them to call it "Luma", or make some other explicit statement to disambiguate the meanings.

-- hide signature --

Alan Robinson

OP (unknown member) Forum Pro • Posts: 16,732
Re: History and Evolution of ISO 12232

alanr0 wrote:

57even wrote:

alanr0 wrote:

57even wrote:

If not, then I guess this article is wrong...

http://dougkerr.net/Pumpkin/articles/SOS_REI.pdf

In a footnote he says "Unfortunately, ISO 12232 neglects to mention this, making its definition of SOS meaningless. In the CIPA standard it is covered only in an annex."

I think he is reading his own interpretation into the standards. The CIPA standard explicitly states that the explanatory annex he refers to is not part of the standard.

So it is a requirement in neither standard, and is not needed if the ISO 12232 test conditions regarding white balance are complied with.

Gotcha.

Interestingly, his article is reference by the Wiki article on ISO.

The Doug Kerr reference by Wikipedia is to the half-stop difference between saturation-based speed and Standard Output Sensitivity.

Easy misunderstanding though, given that they call luminance 'Y' rather than just luminance.

Wikipedia: Luminance: "Not to be confused with Luma (video), Luminescence, or Illuminance"

The ISO standard already uses "Luminance" in the context of "Scene Luminance". If they needed to refer to the Y component, I would expect them to call it "Luma", or make some other explicit statement to disambiguate the meanings.

If gamma adjusted, one should call it lightness, presumably.

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

spider-mario
spider-mario Senior Member • Posts: 1,000
Re: History and Evolution of ISO 12232

57even wrote:

Easy misunderstanding though, given that they call luminance 'Y' rather than just luminance.

For what it’s worth, in the XYZ colorspace, Y is luminance, so there is a precedent for that notation. (Except CIPA then applies sRGB’s non-linearity on top.) Which is why I find it useful to denote luma by Y′ to avoid confusion.

 spider-mario's gear list:spider-mario's gear list
Canon G1 X III Olympus TG-6 Olympus E-M1 II Olympus E-M1 III Canon EOS R6 +14 more
bobn2
bobn2 Forum Pro • Posts: 70,287
Re: DXOMark ISO sensitivity - confused...
5

bclaff wrote:

Slightly Off Topic (OT) but this bit:

Hints at where I got SNR 20 for my Photographic Dynamic Range (PDR) threshold.

As I recall, SNR 5 "OK", SNR 10 "acceptable", SNR 20 "good", and SNR "excellent" appear in various ISO documents.

Which is why I call it an arbitrary choice. You'd need to do better than 'as I recall' and 'various ISO document' if you want to cite a source as justification for a reasoned and non-arbitrary choice. In any case, it's not clear which part of the tonal range the SNR described above refers to, but by implication is it the brightest parts, yet you use it as a threshold for the darkest parts. I can't see that there is much of an argument for an SNR of 20 in the deepest shadows, which is what you are implying if you use it as the lower bound. Also, it is said to apply for images displayed at 70 pixels/cm, which doesn't fit with your CoC normalisation (it would imply looking at a postcard sized image at 25cm).

It's a lucky chance that the very low normalisation resolution and very high lower bound cancel each other out and make 'PDR' look as though it's in the same ballpark as DR.

Note: 12232 explains the derivation of the 10 and 240 figures, as follows:

'The S/N values of 40 for the “first excellent” image and 10 for the “first acceptable” image were determined using subjective experiments performed during the development of this document.'

So it is claiming that they were developed for the purpose of 12232 only, and are not ones that appear in 'various ISO documents'.

-- hide signature --

Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

bobn2
bobn2 Forum Pro • Posts: 70,287
Re: History and Evolution of ISO 12232
3

alanr0 wrote:

ISO 12232:2019 Photography — Digital still cameras — Determination of exposure index, ISO speed ratings, standard output sensitivity, and recommended exposure index

I have not seen the text of this. No substantial changes, as far as I can tell, but I understand from the Wikipedia entry that it defines a wider range of ISO speeds. I also gather that it includes strictures against applying the SOS and REI methods to raw output from cameras.

Not quite, it applies the strictures to the ISO speeds. It doesn't apply them to the exposure indexes, but as below, they make no sense if not in sRGB.

Wikipedia states that SOS applies only to sRGB output, but I do not see this restriction to a single tone curve in the 2006 standard.

The Wikipedia article is a bot confused. but as discussed elsewhere, the SOS makes no sense if sRGB gamma is not applied.

I do not know if there are implications for the DxO saturation-based speed measurement applied to raw output, where a saturation-based approach still seems reasonable.

It is specifically ruled out in 12232:2019, as follows:

ISO speed and ISO speed latitude values may be reported for either scene-referred or output-referred images. ISO speed and ISO speed latitude values shall not be reported for raw images, however, because with raw images processing that affects the values has not been performed.

Sec 6.1, p6.

-- hide signature --

Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

OP (unknown member) Forum Pro • Posts: 16,732
Re: History and Evolution of ISO 12232

bobn2 wrote:

alanr0 wrote:

ISO 12232:2019 Photography — Digital still cameras — Determination of exposure index, ISO speed ratings, standard output sensitivity, and recommended exposure index

I have not seen the text of this. No substantial changes, as far as I can tell, but I understand from the Wikipedia entry that it defines a wider range of ISO speeds. I also gather that it includes strictures against applying the SOS and REI methods to raw output from cameras.

Not quite, it applies the strictures to the ISO speeds. It doesn't apply them to the exposure indexes, but as below, they make no sense if not in sRGB.

Wikipedia states that SOS applies only to sRGB output, but I do not see this restriction to a single tone curve in the 2006 standard.

The Wikipedia article is a bot confused. but as discussed elsewhere, the SOS makes no sense if sRGB gamma is not applied.

I do not know if there are implications for the DxO saturation-based speed measurement applied to raw output, where a saturation-based approach still seems reasonable.

It is specifically ruled out in 12232:2019, as follows:

ISO speed and ISO speed latitude values may be reported for either scene-referred or output-referred images. ISO speed and ISO speed latitude values shall not be reported for raw images, however, because with raw images processing that affects the values has not been performed.

Sec 6.1, p6.

WTH is a 'scene referred image'?

And why should saturation based sensitivity NOT apply to raw data? Or some protocol for it anyway?

For instance - apply standard black and white level offsets and use a greyscale target.

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

Jack Hogan Veteran Member • Posts: 8,064
Re: History and Evolution of ISO 12232

57even wrote:

Interestingly, his article is reference by the Wiki article on ISO.

Don't forget that Kerr (and Nakamura) was writing in 2007, the wording in the standard has been revised quite a bit since then.

Easy misunderstanding though, given that they call luminance 'Y' rather than just luminance.

Well, there is luminance and then there is Luminance. In Bayer CFA digital cameras Y is the result of a compromise, at its simplest expressed as a Compromise Color Matrix from camera 'space' to XYZ (1). Then there is XYZ - and there is adapted XYZ . The sRGB->XYZ matrix shown in the snippet above gets you to XYZ_D65. What was the illuminant at the scene (2)?

You would need at least (1) and (2) to begin to approximate something proportional to Luminance at the scene.

Jack

Jack Hogan Veteran Member • Posts: 8,064
SNR and Perceived Picture Quality
1

bobn2 wrote:

Note: 12232 explains the derivation of the 10 and 240 figures, as follows:

'The S/N values of 40 for the “first excellent” image and 10 for the “first acceptable” image were determined using subjective experiments performed during the development of this document.'

So it is claiming that they were developed for the purpose of 12232 only, and are not ones that appear in 'various ISO documents'.

FWIW those SNR values (10 for acceptable and 40 for excellent) were already in vogue during the early days of cable television in the '80s, perhaps earlier, though I don't remember a TV size and viewing distance being specified

bobn2
bobn2 Forum Pro • Posts: 70,287
Re: SNR and Perceived Picture Quality
4

Jack Hogan wrote:

bobn2 wrote:

Note: 12232 explains the derivation of the 10 and 240 figures, as follows:

'The S/N values of 40 for the “first excellent” image and 10 for the “first acceptable” image were determined using subjective experiments performed during the development of this document.'

So it is claiming that they were developed for the purpose of 12232 only, and are not ones that appear in 'various ISO documents'.

FWIW those SNR values (10 for acceptable and 40 for excellent) were already in vogue during the early days of cable television in the '80s, perhaps earlier, though I don't remember a TV size and viewing distance being specified ;-).

So ISO maybe being less than straightforward also. It does however raise another question about whether it is an appropriate choice for high quality stills photography. I just can't see any reasoning that says that an SNR of 20 is a sensible choice of lower bound here.

-- hide signature --

Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

bobn2
bobn2 Forum Pro • Posts: 70,287
Re: History and Evolution of ISO 12232
1

57even wrote:

bobn2 wrote:

alanr0 wrote:

ISO 12232:2019 Photography — Digital still cameras — Determination of exposure index, ISO speed ratings, standard output sensitivity, and recommended exposure index

I have not seen the text of this. No substantial changes, as far as I can tell, but I understand from the Wikipedia entry that it defines a wider range of ISO speeds. I also gather that it includes strictures against applying the SOS and REI methods to raw output from cameras.

Not quite, it applies the strictures to the ISO speeds. It doesn't apply them to the exposure indexes, but as below, they make no sense if not in sRGB.

Wikipedia states that SOS applies only to sRGB output, but I do not see this restriction to a single tone curve in the 2006 standard.

The Wikipedia article is a bot confused. but as discussed elsewhere, the SOS makes no sense if sRGB gamma is not applied.

I do not know if there are implications for the DxO saturation-based speed measurement applied to raw output, where a saturation-based approach still seems reasonable.

It is specifically ruled out in 12232:2019, as follows:

ISO speed and ISO speed latitude values may be reported for either scene-referred or output-referred images. ISO speed and ISO speed latitude values shall not be reported for raw images, however, because with raw images processing that affects the values has not been performed.

Sec 6.1, p6.

WTH is a 'scene referred image'?

Not sure that I know, typical of committee speak.

And why should saturation based sensitivity NOT apply to raw data? Or some protocol for it anyway?

Just because it doesn't. The philosophy of the digital ISO was to align it with reversal film ISO, that is, that it was centred on visual properties of the final image, not internal properties of the medium. It is rather like basing film ISO on grain size.

For instance - apply standard black and white level offsets and use a greyscale target.

I'm not sure what you mean there.

-- hide signature --

Is it always wrong
for one to have the hots for
Comrade Kim Yo Jong?

Keyboard shortcuts:
FForum MMy threads