Reciprocity failure?

[dmward]

This came up in a discussion the other day.
In the good old days we had to accommodate for reciprocity failure when pushing film.

The question is what about a camera sensor when using the high ISO settings?
Does setting the ISO to H and then using +1 EV comp just set the sensor to the next lower sensitivity or does it work like it would with film?

[Panopeeper]

The question is what about a camera sensor when using the high ISO settings?
Does setting the ISO to H and then using +1 EV comp just set the sensor to the next lower sensitivity or does it work like it would with film?

There is no single answer. Following relates to raw data, not in-camera JPEG.

MFDBs, at least those I analyzed simulate ISO steps (i.e. when the "higher sensitivity" is not achieved by increasing the analog gain) by recording this in metadata; the raw data is not affected at all, it's the task of the raw processor to account for the lower exposure. In other words: the ISO simulation is plain underexposure.

DSLRs do more eyewash and cause lots of damage. The simulated ISO steps are achieved by adjusting the raw data in-camera. This is just like increasing the intensity in raw processing, except that the adjustment of the pixel values before creating the raw data reduces the dynamic range. For example with 12bit raw data depth the maximum pixel value is 4095 (or even less). If the multiplication takes place in-camera, then anything between 2048 and 4095 becomes "clipped", because the new value exceeds the numerical range.

Note, that not only the H settings are fake. For example the Canon 5D2 offers ISO 100-6400 and H1 (12800) and H2 (25600). However, in fact ISO 3200 is the highest real ISO. (Another aspect is, that even though ISO 3200 of the 5D2 is "real", it does not offer any true gain over ISO 1600, i.e. it should not be used, for it reduces the DR by exactly one stop.)

[dmward]

If I understand what you're saying, setting the camera to its highest numeric ISO setting and then using raw conversion to bring up the highlights has no detrimental impact on the blacks since they are at the low end anyway.

I presume this also means there is no benefit from using a minus EV to compensate for "reciprocity failure".

There is no single answer. Following relates to raw data, not in-camera JPEG.

MFDBs, at least those I analyzed simulate ISO steps (i.e. when the "higher sensitivity" is not achieved by increasing the analog gain) by recording this in metadata; the raw data is not affected at all, it's the task of the raw processor to account for the lower exposure. In other words: the ISO simulation is plain underexposure.

DSLRs do more eyewash and cause lots of damage. The simulated ISO steps are achieved by adjusting the raw data in-camera. This is just like increasing the intensity in raw processing, except that the adjustment of the pixel values before creating the raw data reduces the dynamic range. For example with 12bit raw data depth the maximum pixel value is 4095 (or even less). If the multiplication takes place in-camera, then anything between 2048 and 4095 becomes "clipped", because the new value exceeds the numerical range.

Note, that not only the H settings are fake. For example the Canon 5D2 offers ISO 100-6400 and H1 (12800) and H2 (25600). However, in fact ISO 3200 is the highest real ISO. (Another aspect is, that even though ISO 3200 of the 5D2 is "real", it does not offer any true gain over ISO 1600, i.e. it should not be used, for it reduces the DR by exactly one stop.)

[NikoJorj]

Does setting the ISO to H and then using +1 EV comp just set the sensor to the next lower sensitivity or does it work like it would with film?

There is no reciprocity failure as in the previous millenia : exposition remains constant in both cases.

The problem is in other IQ factors : noise and dynamic range among others.
As said by Panopeeper, staying at low ISO increases dynamic range, while higher ISOs settings (particularly "H" ones) decrease dynamic range with no other gain.
However, increasing ISO in the "normal" range (generally 100-1600) does give a gain in term of read noise, which can translate in better image quality (particularly noise in the shadows) in a properly exposed ISO800 image than in a 2EV underexposed ISO200 image.

If you read french, these discussions show examples of the latter :
http://www.chassimages.com/forum/index.php....html#msg629171
http://www.chassimages.com/forum/index.php...ic,24116.0.html

[NikoJorj]

Sorry for the double post... jammed tube on my side.  Please delete this one.

[Panopeeper]

If I understand what you're saying, setting the camera to its highest numeric ISO setting and then using raw conversion to bring up the highlights has no detrimental impact on the blacks since they are at the low end anyway

To be more specific: the detrimental impact on blacks is the reduced exposure. Thus, if you are setting the ISO as high as possible, you are reducing the exposure, and *that* will cause higher noise.

Don't go after ISO but after exposure; select that ISO, which suits with the highest possible exposure you can afford under the given circumstances.

See Connecting Exposure, ISO and Noise