«

Would this kind of thing have any practical applications with
everyday digital images? They seem to use it in more than a few
scientific applications. Some of these packages cost $10,000 or more
so I'm wondering what it can do beyond what current
consumer/photographer image enhancement technologies (i.e, functions
built into Photoshop, etc) we use?
Here's one of the companies that offer it:

http://www.vaytek.com/MicroTomeWin.html

Hi to all,

Deconvolution software is a numerical operation which applys a filter
designed to remove the effect of a convolution. When we photograph an
image, the scene is convolved with the point spread function of the lens in
the camera, and the sensor. This convolution basically blurs the image a
bit. If we had a perfect measure of those two functions, they can be
removed , by deconvolution, and the image would appear to be sharper. This
operation is limited by noise, and , if the sensor is solid state, sampling
of the image by the sensor.

The sharpening algorithms used in photoshop (with the exception of unsharp
masking) are a form of deconvolution, but with an arbitrary estimate of the
blurring function. Unsharp masking is an image dependant operation, while
the deconvolution operations are independent of image data.

Deconvolution routinely occurs in high end point and shoot cameras as well
as in some D-SLRs. The manufacturer has knowledge of the lens and detector
MTF response and they can perform an appropriate inverse filter. When
properly done, in-camera sharpening can be quite effective in removing the
effects of lens and sensor blurring. Unfortunately, as I mentioned before,
this operation is often limited by sensor noise. For this reason, it
sometimes better to do no correction until the final "size" of the image has
been arrived at by all the editing functions. I generally shoot with a
small amount of in camera sharpening turned on when I am at low ISO ratings.
At low light levels, I turn this off and do sharpening operations after the
image has been sized for output.

We also use deconvolution algorithms in our spectral measurement devices
like the i1Pro spectro photometer.

The deconvolution algorithm is almost identical to what you would do for an
image.

It's actually quite routine in all of our digital photo work.

Take care,
Tom L.
--
Tom Lianza
Director of Display and Capture Technologies
GretagMacbeth LLC
3 Industrial Drive
Unit 7&8
Windham, NH 03087
603.681.0315 x232 Tel
603.681.0316 Fax


"Rich" <> wrote in message
news:...
> Would this kind of thing have any practical applications with
> everyday digital images? They seem to use it in more than a few
> scientific applications. Some of these packages cost $10,000 or more
> so I'm wondering what it can do beyond what current
> consumer/photographer image enhancement technologies (i.e, functions
> built into Photoshop, etc) we use?
> Here's one of the companies that offer it:
>
> http://www.vaytek.com/MicroTomeWin.html

tlianza <> wrote:
: Hi to all,
:
: Deconvolution software is a numerical operation which applys a
: filter designed to remove the effect of a convolution. When we
<snip rest of post>

That was just an excellent, excellent post! Thank you **very** much, Tom.


j.

Keep in mind that there is deconvolution, where you have measured your
blur function, and blind deconvolution, where you haven't or couldn't.
Applying a pseudo-inverse filter (an approximation of the Wiener
filter) is fast and effective for deconvolution. The expensive
packages, however, use the Richardson-Lucy method, but I believe this
is too time-consuming for most people. As far as I can see, we are the
only ones offering an easy pseudo-inverse.
Now I think you are talking about blind deconvolution. Most popular is
MatLab with the image processing toolkit. Matlab is very powerful if
you are willing to learn it. We tried to gear our software towards
those who do not want to make a big commitment. And I may be biased,
but I believe our blind deconvolution method works better than theirs.
The program is 'free-to-try' and has several walk-through demos.

Best Regards,
Jim C

>Deconvolution software is a numerical operation which applys a filter
>designed to remove the effect of a convolution.

How does this differ from the DxO software?

On Mon, 6 Mar 2006 07:32:59 -0500, "tlianza" <>
wrote:

>Hi to all,
>
>Deconvolution software is a numerical operation which applys a filter
>designed to remove the effect of a convolution. When we photograph an
>image, the scene is convolved with the point spread function of the lens in
>the camera, and the sensor. This convolution basically blurs the image a
>bit. If we had a perfect measure of those two functions, they can be
>removed , by deconvolution, and the image would appear to be sharper. This
>operation is limited by noise, and , if the sensor is solid state, sampling
>of the image by the sensor.
>

Tom, you just confirmed something I've suspected for a long time.

I was listening to a space mission on NASA TV a few years ago. The
controller asked the astronaut for the serial number of the lens they
used for a particular series of photos. Right then, I figured that
NASA had characterized all the lenses used on that particular mission,
and were able to correct, in post processing, for a particular lens's
aberrations.

Imaging Resource's PMA video interview with DXO further confirmed my
suspicions. In fact, it sounds like DXO has taken this principal to
extremes.

Thanks!

(another) Tom

> We tried to gear our software towards
> those who do not want to make a big commitment. And I may be biased,
> but I believe our blind deconvolution method works better than theirs.
> The program is 'free-to-try' and has several walk-through demos.
>
> Best Regards,
> Jim C

I am assuming that you are referring to "Tria." If so, I'd like to report
that I can't get it to do anything useful ... the program seems very buggy
and turns my computer into a snail. If you were referring to another
program, please correct me. Thanks.

Annika1980 wrote:
>>> Deconvolution software is a numerical operation which applys a
>>> filter designed to remove the effect of a convolution.
>>
>> How does this differ from the DxO software?

DxO have projections for specific lens designs. The software under
discussion takes the process a little further into the unknown by working
without lens specific data. One of the reasons DxO works is that Canon and
Nikon keep pumping out lenses which create faulty images.

The part which facinates me is if the faults of Canon "L" series lenses or
Nikon lenses are so predictable, how come the makers haven't fixed them and
introduced a new model designation costing twice as much? Interesting thing
too is that the only Leica lens DxO cater to is plastic element "vario" on
Panasonic cameras. No Ziess modules either.

Rich wrote:
> Would this kind of thing have any practical applications with
> everyday digital images? They seem to use it in more than a few
> scientific applications. Some of these packages cost $10,000 or more
> so I'm wondering what it can do beyond what current
> consumer/photographer image enhancement technologies (i.e, functions
> built into Photoshop, etc) we use?
There is one vast difference between deconvolution sw for photographic
pictures and the $10,000 scientific package for microscopy you mention: that
and other scientific packages are intended to recover as reliable as possible
3D objects from image data, which is something different than 'beautifying' a
2D image. For example, all blur has to be removed from the image so that
after processing each image plane only contains the objects which are in it.
In other cases objects have to be recovered from noisy 10 photon/pixel data
sets, or images are so large (> 10Gpix) that the software needs to be capable
of running efficiently on large computers. To develop all that including the
necessary support level makes for expensive software.

-- Hans

Speaking of examples, there is an interesting page here:

http://www.bialith.com/Research/BARclockblur.htm

Check out what he allegedly got from the blurred clock image....
Interestingly, he states:

"images shown here are half actual size and jpeg compressed (as such
they cannot be downloaded for attempted repeat of these results)"

Hmmmmm. And there are no other examples. Puzzling. (I can't help
wondering what an averaged version of all those attempts would look
like, but I can't be bothered..) In some ways, these images
demonstrate my point - s/he gets some fascinating, almost unbelievable
results, but look at all the artefacts..