[filmscanners] Re: PS sharpening

["Anthony Atkielski" <anthony@atkielski.com>]

David writes:

> Just to clarify here: the sharpening with
> radius of 4.9 pixels or so is applied _before_
> downsampling by 500%, obviously. Right?

Yes, it would have to be, otherwise the information it needs would be gone.

However, I haven't actually done this, so I'm not sure of the details.

It seems that, from a mathematical standpoint, there should be a one-step
equivalent of the multiple-step process that I use, but I've always been too
lazy to try to figure it out.  Additionally, I suspect that any one-step
process would require some degree of calculation for each image and each
downsample ratio, and I'm not really in the mood to do that each time I
downsample.  The multiple-step process is easier and seems to give the same
results.

> As I understand it, there should be N + 1
> sharpening operations for N downsampling
> opertions.

The other way around:  N-1 unsharp masks for N downsamples, unless the last
downsample is very close to 2x itself (try it both ways and pick whichever
looks better for the last step).

> In some sense, the first N sharpening operations
> have a different purpose than the last: they're
> to make sure the downsampling retains the detail
> (and contrast) you want.

Yes.  By unsharp masking after each downsample, you exaggerate detail.  The
traces of this exaggeration survive into the next step.  The net result
after several steps is that details that normally would have gone away in
the downsampling still have left tiny traces in the final image.
Technically, the image is flawed because of this, because the details are
exaggerated far more than would be mathematically appropriate--but since the
image is being seen by human eyes, this exaggerated detail is exactly what
is needed to give an impression of greater detail and sharpness.

> The last sharpening is to make the final
> image look good.

Exactly.  The intermediate unsharp masks just help to carry important detail
through the process; only the last unsharp mask is purely aesthetic.  Or at
least that is my opinion; like I said, I've not tried to come up with a
mathematical proof.

Try downsampling through a 100:1 ratio in steps, and then in one pass, and
you'll see that doing it in steps gives you a final result that looks like a
tiny, sharp version of the original, whereas a single step just produces a
blur.



----------------------------------------------------------------------------------------
Unsubscribe by mail to listserver@halftone.co.uk, with 'unsubscribe 
filmscanners'
or 'unsubscribe filmscanners_digest' (as appropriate) in the message title or 
body