Balancing Hi ISO and Exposure decisions.

A while back we discussed the effects on noise if comparing images correctly exposed with images under exposure & subsequently exposure adjusted in post processing. e.g ISO6400 with exp comp -1.0EV compared to ISO12600 at 0.0 EV exp compensation.

I found some related article which may be of interest.
The tests:-
http://ishootshows.com/2009/01/26/hi...sh-processing/
The conclusion and theory:-
http://ishootshows.com/2009/01/28/pu...nd-unity-gain/

From the writers point of view & using Nikon D3 he suggests pushing an under exposed ISO6400 during post processing can be judged as better than exposing correctly on ISO 12800. He makes only a delta 1 stop evaluation however.

Whilst I do not particulary agree or disagree, what I found interesting was his remarks on "unity gain" issues - which I have to admit struggling to understand fully. Nevertheless it is an interesting theory and I would appreciate any pointers to similar so called unity gain ISO levels by camera model. I use a D700 and wonder if sticking to a maximum of 6400ISO is valid.

I suspect that Nikon do their high-ISO captures in a similar way to Canon in that anything in the "extended" category is simply (and essentially) a digital manipulation of data, which I think makes evaluation of images into this area a different kettle of fish to those taken in the more conventional (ampified) ISO bands.

Many are now starting to comment that some of these extreme ISO settings appear to be achieving their "good" results more by using high-ISO noise reduction techniques than through advances in sensor technology.

It seemed to me that the results of that test were relatively inconclusive (I personally would have gone for the B Image).

Even if the higher ISO value is real (i.e. it is not software fake ISO), the improvement in SNR by rising ISO beyond some particular value (Emil Martinec claims ISO1600 would be a proper value for most cameras) becomes negligible, so even if the RAW file results underexposed there will be no improvement for rising ISO.

The reason for this is that from some ISO value and higher, noise introduced in the last stage of the RAW generation (AD converter basically) becomes much smaller than the noise output from the ISO amplifier because of the high gain used, so SNR does not improve anymore since most of the noise finally saved in the RAW data comes from the first stages.

And using a lower ISO will always help preserve better the highlights in case some part of an overall dark scene would have some highly lighted area.

Emil explains this, and also shows some disagreement with the 'unity gain' theory here.

BR

Keeping in mind though that as far as we can tell, Canon (for one) appear to be doing high-ISO noise reduction post A/D conversion (assuming the option is enabled) - not sure if this changes anything.

On my D90 I have the option of high iso Noise Reduction ON or OFF, some opinions are to leave it off as it tends to create more problems than it cures.