View Single Post
  #17  
Old September 15th 14, 08:12 AM posted to rec.photo.digital
Martin Brown
external usenet poster
 
Posts: 821
Default Lenses and sharpening

On 13/09/2014 08:23, Alfred Molon wrote:

Sometimes a soft lens can be very effectively compensated by some
unsharp mask in post processing and you get a sharp, natural looking
image.


For this to work well some very restrictive conditions have to be met -
notably that the entire image is uniformly blurred by a symmetrical and
spatially invariant point spread function (or one which varies slightly
and in a very predictable way with distance from the optic axis).

But sometimes no matter how much sharpening you apply or what parameters
you choose, you get that unnatural, "sharpened" look.


Anywhere that was already sharper to begin with than the presumed blur
will end up looking unnatural with ringing artefacts around it.

It probably depends on the unsharpness of the lens, its (spatial)
frequency response or whether the sharpness is caused by the lens glass
itself (i.e. lens not being sharp enough), inaccurate focus or some
motion blur.


You can only take out motion blur by modelling it. It can be done for a
price provided there is something in the image to play guess the
blurring function from. Used in car number plate recognition sometimes.

http://www.maxent.co.uk/example_1.htm

"Blind" deconvolution can also be done on larger images provided that
the blurring function is pretty much the same over the entire image.

For instance I have a 70-300 lens which at the tele end generates a bit
soft images, which however respond well to unsharp masking in post-
processing. But that's not the case for the another lens I have (a mid-
range one).


A lens which scatters a bit of light into a halo around everything migth
well tune up OK with a bit of unsharp masking.

Has somebody analysed this (i.e. how to best sharpen an image, what
unsharpness can be eliminated in post-processing)? Is there perhaps some
web page with details?

Astronomers have the most detailed instructions on how to deconvolve
images to obtain the maximum information about the night sky from finite
sized and imperfect instruments. They have one big advantage over
conventional photography in that their targets are all at infinity.

A rough rule of thumb is that at the highest signal to noise you can get
3x the spatial resolution on the highlights and brign the noise under
better control in the mid range. The price you pay is some unfamiliar
artefacts in the image which can be a nuisance in medical diagnostics
where practitioners are very familiar with the quirks of the standard
linear inverse methods for image reconstruction.

Depth of field issues make life very difficult for normal photography.

--
Regards,
Martin Brown