A couple of people have mentioned this, and I agree with them completely. The huge problem here is not really any of the above except for that image is waaaay too underexposed as shot. It looks like it needs at least another stop and a half of exposure, more likely two.
Well put.
More exposure yields data with more information content. More information content increases post-production rendering flexibility. This is the case right up to when over exposure exceeds the photo sites' full-well capacity and, or exposure combined with gain (camera ISO setting) causes analog signal clipping during digitization. The former happens at the camera's native ISO and the latter occurs after the shutter closes and the camera ISO setting is above the native setting.
...
Being able to manipulate images digitally is no justification for not getting your exposure right in camera. ...
It turns out cameras with ISO-invariant sensor assembly designs do not require "getting the exposure right in-camera". These cameras have very low read (electronic) noise levels up to quite high camera ISO settings (1600 or more). This means image noise is dominated by photon noise (noise inherent to converting electromagnetic radiation to photoelectrons). The result is essentially identical signal-to-noise ratios for images made at different camera ISO settings using identical shutter times and apertures. Increasing signal gain (using higher camera ISO settings) does not improve perceived image noise levels.
However, cameras with ISO-invariant sensor assembly designs
do require "getting the exposure right in-camera". Now "getting the exposure right" means maximizing exposure. Exposure is maximized by using the longest practical shutter time and widest aperture that does not overexpose or clip important highlight regions. This approach is only useful for raw files. The disadvantage is in-camera image review (JPEGs) could be compromised due to dark images.
Here are some
comparisons of shadow region noise vs. camera ISO setting for the M9, M240 and M10. These noise level estimates are computed using unrendered raw-file data.
Note the difference between the M9, M240 compared to the M10. For the M9 and M240 analog signal gain (in-camera ISO setting) improves shadow region rendering. Read noise levels are important. The M10 has essentially the same (within 0.1 EV) shadow region noise between ISO 400 and 1600 while the M9 and M240 data shows shadow region noise improvement over the same range. The M10 exhibits very little (if any) dependence on read noise levels.
One shocking example of ISO invariance is the Sony Alpha 7s II. The shadow region noise level (
data) is ISO-invariant from ISO 2000 through ISO 80000!
These blog posts provide more details.
ISO Dependent
ISO Invariant