It is just that I consider capture sharpening (preferably by deconvolution) as an essential part of the process, especially if there is an AA filter. I understand it is difficult to include it in a comparative test and not cause more controversy.
Hi Francisco,
I agree that capture sharpening is an essential part of the image formation process. It's however not necessary in some situations, and it may be detrimental in others. I think it's useful to distinguish between two categories of questions:
1) How will lens A perform compared to lens B on my one camera?; and
2) How will lens A perform on two different cameras with varying AA strengths and most other things equal?
In the first case a metric to help with buying decisions should imo concentrate on the hardware only. As a (amateur landscape) photographer I consider the job to capture the best spatial information possible at the scene so that such information can be later processed to provide the most pleasing results when viewed on the final display medium as desired. The better the information captured, the better the final result. The best spatial information captured is obtained by using the best hardware available, so what I am interested in is the objective performance of the hardware in a camera system when selecting equipment - as opposed to subjective post processing workflow results (including demosaicing and sharpening algorithms with arbitrary parameters) which can always (and mostly will) be added later to obtain the desired effect. Perhaps this leaves open a question of diminishing returns past a certain point, but that's one I am willing to deal with.
In the second case I can see the rationale for applying capture sharpening before spatial resolution measurements because it is true that one may be able to 'restore' some of the effects of a stronger AA filter or other subsystem through the judicious application of advanced sharpening algorithms. On the other hand things get very fuzzy and subjective very quickly because perception enters the equation prominently. Plus we all know that typical sharpness metrics are very sensitive to acutance, whether that be due to the real performance of the system or introduced artificially by an overzelous operator. It doesn't take much.
Take for instance Photozone.de which opens images for MTF measurements in ACR/LR with sharpening at default, what most observers would consider mild capture sharpening. Edge Spread Function profiles so generated show overshoots and undershoots in what in the physical world is actually a monotonically increasing S-shaped curve. Even that is too much. Perhaps a reasonable criterion for how much capture sharpening to apply should be that physical limits not be exceeded. Easier to say than to do. Or perhaps the operator should sharpen both images to the best of his subjective evaluation/abilities following a strict set of criteria. Again... And this says nothing of the amplified noise and introduced artifacts that such an approach would entail.
Too many balls in the air for my taste. So while I see the rationale for the application of capture sharpening before measuring spatial resolution for some uses, I have a really hard time figuring out how such measurements can be practically useful by themselves. That's why I tend to start at Lenstip.com (mostly hardware), move on to Photozone.de (capture sharpened) before reaching DxO (perceptual milk shake) and then drawing my own Jack-Pix purchasing conclusions :-)