Blind deconvolution such as Lucy-Richardson (FM) seems to do quite a
good job if over correction with "image ringing" and other artifacts
are avoided despite total ignorance of the real PSF.
I thought some of my earlier posts a few years ago may actually have
resulted in capturing Moose's interest. As I understand it, most assume
the Point Spread Function (PSF) is gaussian (that which blurred th
image in the first place) and converge on the maximum likelihood
estimation after an iterative process. (focus my Hubble, please)
Linear motion is another possible PSF and I think will be correctable
in the next PS CS version. DXO has attempted a light version of what
Moose thought would be nice to do.
Back 5 years ago or so the tech support at DXO was less guarded in what
they would reveal. I very much doubt their lens specific profiles have
true full wave solution PSFbut apply
blind deconvolution in a weighted manner where the lens has the most
need for it. Perhaps the "global slider" is a radius adjustment and the
"detail" somewhat controls the iterations.
(Just a guess) The focal distance specific correction of CA and
geometric distortion included in the lens profile as well as the noise
profiles are a plus. Too bad it doesn't quite
live up to the concept but is worth using in some circumstances and
exporting a tiff to PS.
Keeping the convoluted in deconvolution, Mike
Convolution ... Deconvolution would be changing between frequency and
linear dimensions/ features. By measuring the frequency response of the
system you can adjust/improve the output without being concerned with
the
individual parts of the system.
Jeff
On Feb 7, 2012 10:04 AM, "Moose" <olymoose@xxxxxxxxx> wrote:
> On 2/7/2012 8:47 AM, Chris Barker wrote:
> > Thanks for that appreciation, Moose, but what is deconvolution -- in
> layman's terms? I've had a search on the Internet, but the
explanations
> mean nothing to me.
>
> Non-scientific, possibly/probably therefore not entirely accurate, but
> indicative in a possibly useful way:
>
> When light is focused through a lens, the various imperfection in the
lens
> convolute it, "intricately fold, twist, or coil"
>
> If the characteristics of the lens and the focal and image distances
are
> mathematically defined, it is theoretically
> possible to un-fold, un-spindle and un-mutilate the resultant image
into
> what it would have been if formed with a
> perfect lens.
>
> In practical terms, the lens can't be perfectly known and described in
> math, but the results of deconvolution may be
> startlingly good.
>
> At the level I'm talking about, with Focus Magic and other consumer
tools,
> we have generic applications of deconvolution
> that make lots of simplifying assumptions about the lens that may have
> formed the image and allow specification of
> simple input(s) and choice of input criteria based on viewing a small
> sample of the image at 100%.
>
> Whilst far from the potential of custom tailored deconvolution, these
> simple versions still accomplish semi-magical
> transformations in many cases. FM, in this case, seems particularly
suited
> to images from the S100, for whatever reasons
> of lens itself and sensor system. Oddly, this seems to be true across
the
> zoom range.
>
> Deconvolution is rather processor intensive. My speculation was that
> someone, one of these days, will have an in-camera
> processor with the power to do custom, in-camera deconvolution
specific to
> the lens, focal length and focal distance. I
> suspect the results will be spectacular.
>
> In the meantime, I suspect that some enterprising geek will eventually
> come up with a way to create lens profiles for
> any lens and read focal distance from EXIF to make the corrections in
a
> post processing app.
>
> Looking at the corners of JPEG vs Raw files from the S100 in DPP,
Canon's
> own Raw converter, I see some quite amazing
> corrections going on. Smeary, distorted stuff becomes clear. I don't
think
> is is deconvolution, in the full sense, but
> I'm impressed. How do I work that into my work flow when corner
details
> matters to me?
>
> Convoluted Post Moose
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|