Underexposure vs raising ISO, a practical test

hjulenissen wrote:

Yes you are right there is some clipping in the ISO 1600 file green/blue channels (LCD display) and green channel parts of the white bricks. Like you say, I think that is a benefit of under-exposing.

I am sceptical that any sane processing can hide the obnoxious shadow noise in the ISO100 file while keeping detail. Beside from manual pixel editing (where an image of someones cat can, in principle, be turned into Mona Lisa):

Crazy Reply Editor won't let me delete that gray bar.

Anyway, the vertical banding part of the noise is not something that should be difficult to correct.  There are two components to it; a soft periodic banding, and a harsh random banding.  The harsh random part is the same in every frame.  However, it does not disappear with Long Exposure Noise Reduction, but it does if subtract a black frame yourself, so I assume that it actually an error created by the camera firmware, and not really a noise per se.  The periodic banding is not fixed, however, but if the camera had enough masked pixels across the top of the frame, you could use that data to partially correct both components, but Canon, somehow, only has 31 rows of what look like masked pixels in the RAW file, and only 13 of the look like they are really masked pixels, when the period of the periodic part is an integer number of pixels however, as it is here, you can use columns 0, 8, 16, 24, etc, to correct the columns in modulo.  I tried it once a while back when I first got my 7D, and it reduced the periodic quite a bit, but not completely as some periods seemed to be significantly above and below their modulo siblings.  Having 100 to 200 real masked rows of pixels would make this so much easier to correct properly, because the vertical banding in the 7D is extremely consistent from the top of the frame to the bottom.