Pentax 645 on DxO mark

[eronald]

To address your example, this is a good example where a lack of understanding led to casting suspicion on the data.  The answer is actually quite interesting:  the value recorded for all ISO settings in those Hasselblads is the same is because this camera is not providing hardware based signal gain.  DxO's information is telling you that the ISO settings you select on the camera are applied by software, after the capture is made.

As Bernard might say, this explains why at ISO 800 highlight recovery is so good

Edmund

[BJL]

I have not found a detailed definition of their ISO measurement either except a description that they are fully compliant with ISO standards, which would mean that the ISO is measured the same way it was in the film days.

Not really: remember that ISO just refers to the International Organization for Standardization which publishes many standards, including several related to film and sensors, and the ISO standard for film speed is not the same thing as the so-called ISO setting on a digital camera, though they have something in common.

1. The "ISO scale" on a digital camera should relate to Exposure Index, which measures what combinations of subject illumination, shutter speed, and aperture ratio will give a suitable "middle gray" response from the camera. This really should be called exposure index, but the ambiguous and confusing name "ISO" is entrenched.

2. The ISO (formerly ASA) speed rating of a film is roughly the maximum exposure index that gives results meeting a certain standard of shadow handling and such: it is roughly a maximum safe exposure index, based roughly on how well subjects four stops below middle gray are handled.

By the way, there is yet another ISO standard definition worth mention:

3. There is also an ISO standard for "base speed" or roughly the minimum exposure index giving satisfactory highlight handling; based on having about three stops between mid-tone exposure level and blown highlights. Quite different from ISO film speed.


Loosely, 1. is about specifying units of measurement (like seconds) while 2. and 3. use those units to measure two quite different characteristics of the "light sensitive medium" (the minimum and maximum acceptable exposure times in those units of seconds under certain conditions on illumination and f-stop.)

[jduncan]

The answer is actually quite interesting:  the value recorded for all ISO settings in those Hasselblads is the same is because this camera is not providing hardware based signal gain.  DxO's information is telling you that the ISO settings you select on the camera are applied by software, after the capture is made.

Excelent explanation.
It was my first guess when I saw the data. Of course is I am a CS by trying, and this cameras are electronic devices, my bread and butter.
For some photographers its more like me in a convention about car engines for me: I  know the general stuff, and the physics ,  but my intuition and training is not in that area. None of that is any help in driving a car by the way.

Maybe DxO should make a more Photographer friendly, easy to find explanation?.
The lack of gain (signal amplification) could be compared to the difference between  fast film and pushing slow film.

[ondebanks]

<snip> hich would mean that the ISO is measured the same way it was in the film days.

As Bradley & BJL have explained, that's not how digital ISO is measured.

But just imagine if it was measured using the same basis as film was: minimum exposure required to register some threshold density above base fog, would translate to minimum exposure required to register some threshold signal to noise at a pixel (say, 3 sigma, the normal minimum threshold in science). Instead of all cameras having base ISOs within a factor of 2 or 3 of each other, P&S digicams and phone cams with tiny pixels would be something like ISO 50 while the likes of a Nikon D3 would be something like ISO 3200 - and I'm talking about base ISO. This would correlate roughly with Roger Clark's concepts of Unity Gain ISO (considering signal shot noise only) or Low Light Sensitivity Factor (considering readnoise as well).

That would transform the digital camera scene and the public's perspectives of it! Instead of pursuing largely useless metrics like megapixel count, they would judge on a metric that really does tell you something significant about the capabilities of the sensor, and in what shooting situations it can be expected to perform well or poorly.

[madmanchan]

The ISO 12232 standard discusses various methods for characterizing the final output values and how they relate to reported values (e.g., ISO 100, REI). It says nothing about how the raw data itself is recorded in relation to the reported values.

[bjanes]

As Bradley & BJL have explained, that's not how digital ISO is measured.

But just imagine if it was measured using the same basis as film was: minimum exposure required to register some threshold density above base fog, would translate to minimum exposure required to register some threshold signal to noise at a pixel (say, 3 sigma, the normal minimum threshold in science). Instead of all cameras having base ISOs within a factor of 2 or 3 of each other, P&S digicams and phone cams with tiny pixels would be something like ISO 50 while the likes of a Nikon D3 would be something like ISO 3200 - and I'm talking about base ISO. This would correlate roughly with Roger Clark's concepts of Unity Gain ISO (considering signal shot noise only) or Low Light Sensitivity Factor (considering readnoise as well).

Wikipedia has a pretty good explanation of the ISO standard 12232:2006 that is used to rate digital camera sensitivity. DXO uses the Saturation-based speed, wherein exposure settings based on a standard light meter and an 18-percent reflective surface will result in an image with a grey level of 18%/√2 = 12.7% of saturation. This allows a half stop of headroom for the highlights. If you want highlights to be at saturation (as in exposing fully to the right) you have to add 1/2 stop of exposure as determined by a gray card reading. This hasn't changed from the film era as explained by Thom Hogan.

The ISO also has a SOS rating in which 18% saturation of the sensor corresponds to an 8 bit RGB value of 118 in a gamma 2.2 encoded output space. Strictly speaking this rating does not apply to raw data, since the standard specifies a gamma 2.2 space for a rendered file. However, the concept is often applied for raw data (see Bill Claff, who has compiled ratings for various Nikon cameras). The JPEG engine in most digital cameras and most raw converters apply a tone curve so that the value in the rendered file may well not be 118. To determine the sensor saturation, you have to look at the raw linear data with a program that allows this (DCRaw and Iris are two such programs). For example, a gray card (or for that matter a white card) exposed according to the light meter reading with my Nikon D3 gives a 12 bit value in the green channel of 497, which corresponds to 497/4095 or 12.1% saturation, which is in accord with the ISO 12232:2006 saturation standard. However, the JPEG engine with the Standard Picture Control renders this with a pixel value of 150 in sRGB. This represents a hot tone curve and the rendered image appears overexposed. You really have to know your camera and the rendering hardware.

The standard does allow for a noise based speed rating much as you propose, but this is seldom used. With digital, we really want to expose for the highlights (exposure to the right).

[BJL]

I agree with ondebanks except on one word, "base".

... P&S digicams and phone cams with tiny pixels would be something like ISO 50 while the likes of a Nikon D3 would be something like ISO 3200 - and I'm talking about base ISO.

Not base ISO. That word "base" in ISO spec's refers to the minimum exposure index giving adequate highlight handling, and it rather naturally does not depend much on pixel size. That ISO 3200 for the D3 would be the maximum exposure index for adequate shadow/noise handling, which I propose calling the "film-like ISO speed". In film exposure/processing terms, the base ISO speed is how far you can safely pull; the shadow/noise based speed limit is how far you can safely push.

[JDG]

For clarity, DxO Mark references ISO Standard 12232 for ISO and uses the saturation method defined therein.  It's also worth noting that ISO measurements are done without a lens mounted to ensure what is being measured is in fact sensor sensitivity, and is not biased by the transmission characteristics of particular optics.

http://www.dxomark.com/index.php/eng/Learn-more/Understanding-DxOMark-Database/Measurements/ISO-sensitivity

And just in case anyone thought Bernard was making this stuff up (from the above link):

"As tests show, the ISO settings reported by camera manufacturers can differ significantly from measured ISO in RAW. This difference stems from design choices, in particular the choice to keep some “headroom” to avoid saturation in the higher exposures to make it possible to recover from blown highlights."

For the sake of completeness, it is worth noting that ISO standard #12232 is not without its critics:
http://doug.kerr.home.att.net/pumpkin/index.htm#ISO-DR (link seems to be down--I'll try to find another)

but despite Mr. Kerr's thoughtful analysis, I don't think it is reasonable to make the claim that DxO's methodology is arbitrary.

To address your example, this is a good example where a lack of understanding led to casting suspicion on the data, instead of, perhaps, asking for clarification.  The answer is actually quite interesting:  the value recorded for all ISO settings in those Hasselblads is the same is because this camera is not providing hardware based signal gain.  DxO's information is telling you that the ISO settings you select on the camera are applied by software, after the capture is made.

I appreciate your inclusion of the link I was looking for.. just couldn't find it before some reason.  I didn't mean to suggest their methods were arbitrary, but simply that depending on their methodology they might not tell the whole story.

[ondebanks]

Thanks for the clarifications & corrections, bjanes & BJL.

The standard does allow for a noise based speed rating much as you propose, but this is seldom used. With digital, we really want to expose for the highlights (exposure to the right).

There are a few domains away from the mainstream of digital imaging, like deep-sky astrophotography, where we don't have this luxury of ETTR, and a noise based rating would be far more useful. I came to this realisation too late in my own learning process, or more accurately I let my MF-loving heart override my S/N-calculating head, when I bought a used MFDB instead of a 5DII. Surely the bigger pixels and larger total light-gathering surface area would let the MFDB hold its own against the 5DII's lower readnoise? Alas, no. I should have heeded the ISO ratings, and done up those S/N spreadsheet calculations sooner! Still, the MFDB is better at just about all the daytime stuff I like to shoot, so it's not a tragedy; but definitely a disappointment.

[madmanchan]

Yes, the ISO 12232 standard has different reporting methods (saturation based, noise based, recommended exposure index), which are reflected in the revised EXIF 2.3 spec.

Again, this has nothing to do with shooting raw, though ...