On 11/5/2014 1:35 PM, Mike Gordon via olympus wrote:
Convoluted Moose writes:
... That reminds me. I still haven't re-compared DxO's lens/sensor specific deconvolution to FM's generic version.
OH well, another day.
Yes, Yes! Please do the latter when it suits. The new version tweaks the defaults to hit the image with more
deconvolution than before using a slightly improved algorithm.
Would be sure the sliders are adjusted to under-do the image and port to PS and tweak further there. Can use FM or USM
after that as well.
Maybe when you and Chuck finish hitting me over the head with theory, but
without practical examples? :-)
...
Now on the other diffraction issues. I think I am tickled being channeled but am not sure. I think I should re-post
a few things discussed over months a while ago.
About a year ago my working hypothesis was that even true point-spread deconvolution will NOT work all that well to
fix blur due diffraction from a small aperture.
I just don't know. But remember, my approach is the opposite. I take pics, process them and base my parameters for
shooting on those results. From that viewpoint, it matters not which lens/sensor effects degrade the images, only how
well the problems may be corrected.
Recall from Dr. D's link that on FT sensor by F11 max resolution is down to
about 4MP overall and 3MP
with red light.
http://www.luminous-landscape.com/tutorials/resolution.shtml (see table 3)
I still can't read these things very well. Perhaps it's because they depend on
contrast?
Unlike diffraction blur ,Gaussian blur "leaks" enough high frequency
information into the image that it can be boosted and reconstructed
(indeed, a gaussian is a very special extremum case that results in max
blur for the least loss of high frequency detail)
I'm not sure how this applies. If Gaussian blur isn't part of what the lens
does, why do I need to know this?
found the below images--not perfect but good enough
original image:
http://praetoriusphoto.images.s3.amazonaws.com/fmforums/20120322_deconvolution/sinsweep.jpeg
added diffraction:
http://praetoriusphoto.images.s3.amazonaws.com/fmforums/20120322_deconvolution/diffracted_sweep.jpg
note how detail frequencies above the cutoff are almost completely wiped out.
See? This is where I go a little nuts with the pure theory. Manufactured images with no indication at all of what they
might represent in scale on a sensor or in aperture.
OMG!!! Awful, just awful what diffraction is doing to my images. Oooooohh
(Swoons)
But wait! How about a little exercise? Just as theoretical, in a way, but the processing is real.
<http://www.moosemystic.net/Gallery/tech/DiffractionContrast/diffracted_sweep.htm>
What if - - my sensor, and system as a whole, with some near perfect lens,
can't resolve all that detail anyway?
- Off with the end of the chart!
What if - - I'm capable of processing the image to recover detail?
- Oh! Never mind.
I of course have no idea where such a chart might intersect with my system resolution. Nor does the fellow who made this
scary, but completely ungrounded, example.
One problem with bringing it into the real world is that there are other things going on with the lens at the same time.
Another is that I seldom photograph sine waves this way. ;-) You may have noticed that this particular processing also
squared up the lines, eliminating the soft horizontal edges at the big end. Depending on actual scale on sensor, this
may make everything appear sharper at the pixel level, or make it look all weird at a larger level. I could have used a
graduated mask, and you wouldn't see it.
My guess, based on way too much pixel peeping when processing, is that it starts to become significant around f8 and too
severe to really correct past f11.
Gassian blur added:
http://praetoriusphoto.images.s3.amazonaws.com/fmforums/20120322_deconvolution/gaussian_sweep.jpg
note gradual reduciton of high freq info rather than cut-off.
HOWEVER, form a practical vantage point, Moose has always said that stopping
down for dof if required for
the image even past the diffraction threshold should usually be done
though maximal sharpness in the in focus areas may be slightly
compromised. When aperture bracketing when I was previously more
nervous going way past the sweet spot for a lens, empirically the dof
gain almost always offset
the modest diffraction softening (as I don my I agree with
Moose T--shirt) unless the image is turned to mush at very small
apertures. Spotted this more exacting review of the situation :
I always look for the lesser of the two Weevils.* The reason the following examples tend to agree with what I've been
saying for ages is simple; they got out of their heads and did some experimentation. It's not for nothing that I spent
time doing stuff like taking pics on a tripod of very deep, highly detailed subjects (fleabane, dandelion heads, etc.)
at various apertures and focal depths. The 135/4.5, for example, is really pretty decent, quite usable, to f22 on 35 mm
film. (Rather like I've found for f11 on 4/3.) At f32, it's fallen off the edge. Resolution charts don't tell you the
right things!
http://www.lensrentals.com/blog/2013/03/overcoming-my-fentekaphobia (look at
this link if nothing else!)
To quote Roger at Lensrentals from the link:
"The message I took away, though, is that diffraction softening is real, it occurs where it is supposed to, but it’s
really not as severe
as I had thought. Even on the D800 resolution is as high, or higher, at f/16 than it was at f/2.8. At f/11 the
resolution is as good, or
better, than at f/4. And at both f/11 and f/16 resolution is clearly higher than it was wide open. Perhaps the
diffraction monster’s teeth
AREN'T as long and wicked as I thought."
Toothless weevil? Good job. I like that he figured out that processing works differently at the extremes of the aperture
range. That's really a help sometimes.
Now look at link at post 66:
http://www.luminous-landscape.com/forum/index.php?topic=45038.60
Yikes, R-L deconvolution didn't do half bad ! Admittingly he used a
perfect PSF but still much of the compromised image data should have
been gone and what is there is distributed over more energies requiring
much more intensive processing at diminishing returns to recover.
So deconvolution CAN improve diffraction softened images ( Canyon's DLO uses an exact PSF) despite total loss of some
high freq image data.
How can that be??????
Again, the problem of living in one's head, spinning theories around. The real monster under the bed is just a big dust
bunny. :-)
Deconvolution sharpening can (with non-determinate PSF's) do increase
detail CONTRAST, though you can't really increase the maximum detail
frequency. It gets further confusing in that unfortunately the concept
of resolution has to be very tightly coupled to "contrast", and often
you use MTF50 contrast to get a "resolution" number.
No, 'You' do, not I.
The point of real detail extinction is significantly higher than that.
Ah, grasshopper (... to move on in Insectivora), now you may start to understand why MTF charts don't seem as useful to
me as everyone else seems to find them. (BTW, same applies to all those shots of hi contrast lines on which old lens
reviews relied.)
So FM or DXO Optics Pro away on diffraction softened images and go ahead and push the aperture up past the diffraction
limited thresholds.
Keeping the convoluted in deconvolution, Mike
Convolusalated Moose
* But don't remember the joke that line comes from. Yeah, I know, it's in a
movie, but I know it predates that.
--
What if the Hokey Pokey *IS* what it's all about?
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|
