Unmasked Tempel 1

jim wrote:
> 
> Jerry Avins wrote:
> 
> 
>> That's a sharpening mask, not unsharpening. A typo, I presume?
>>
> 
> 
> No I meant what I said. An unsharp filter is when you subtract a blurred
> version from the original:
>    g(x,y) = f(x,y) - f'(x,y) 
> 	where f'(x,y) is a blurred version of f(x,y)

Yes. I've done that myself. He uses, if the web site is correct, a 
peaking filter. He claims that a net gain of 1/9 rather than the typical 
unity is somehow beneficial, but that he doesn't know why.

> 	The smoothing filter here is just a box car filter which is a bit
> crude. 

Look at the difference between the central coefficient and all the 
others. Do you still call it a boxcar?

>       But if you apply the smoothing (the box-car part) with many
> iterations it becomes more gaussian like. And if you give the user
> control over the amount of that the smoothed part gets subtracted from
> the original you have an interactive de-blurring tool that could produce
> results like he is showing. 

He gives no such control his only variable is the number of passes with 
the only kernel he supports.

> 	This is a standard technique for improving blurry images that has been
> used by photographers since before computers were around. I'm guessing
> what he is claiming as unique is the method of creating the blurred
> version -> f'(x,y)

I must have missed that part. Where does he write about creating a 
blurred version by any method?

   ...

>> There is an anomaly in Tontoko's images. Each pass of the filter kernel
>> leaves garbage at the edges of the image, yet none is visible in the
>> sharpened images. ...

No comment on that?

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Rune Allnor wrote:

> 
> jim skrev:
>> Rune Allnor wrote:
>> \
>>
>> > Just to elaborate: No one dispute that the processed images
>> > look nicer than the originals.
>>
>> Actually they look extremely nice. If you apply the simple mask operator
>> that you are accusing him of applying to his image you don't get
>> anything even close to as nice as his result. He is obviously doing
>> something a lot more sophisticated than his detractors imagine. Its not
>> just a simple unsharp mask. What it is (or how its implemented) is not
>> at all clear from his descriptions.
> 
> I don't know or understand what he is doing; that's why I asked the
> questions in the first place and why I followed up by asking if he
> could
> post the originals. The description of what he calls a "focus
> corrector"
> is, as far as I can see, merely a mask operator implemented in floating
> 
> point arithmetics.
> 
> Either this guy is doing something other or more than what he
> says he does, or the fixed/floating point issue is way more important
> than at least I would imagine at the outset.
> 
> Rune

It seems from the description on the OPs web-page

        http://139.134.5.123/tiddler2/c22508/focus.htm

that he is not doing a simple convolution with the mask. Mainly he claims,
that it is not the central pixel thats being modified, but the surrounding
pixels of the central one (i.e., the red coefficients). I don't see how
thats going to work. 

In addition, if you follow the 'mathematical note'-link you find that he
claims to be inverting some matrix based on geometrical optics that usually
yields convolution.

-- 
Jani Huhtanen

Jerry Avins wrote:
> 
> jim wrote:
> >
> > Jerry Avins wrote:
> >
> >
> >> That's a sharpening mask, not unsharpening. A typo, I presume?
> >>
> >
> >
> > No I meant what I said. An unsharp filter is when you subtract a blurred
> > version from the original:
> >    g(x,y) = f(x,y) - f'(x,y)
> >       where f'(x,y) is a blurred version of f(x,y)
> 
> Yes. I've done that myself. He uses, if the web site is correct, a
> peaking filter. He claims that a net gain of 1/9 rather than the typical
> unity is somehow beneficial, but that he doesn't know why.

You are so hung up on your naming conventions for filters you can't see
the forest for the trees.


> 
> >       The smoothing filter here is just a box car filter which is a bit
> > crude.
> 
> Look at the difference between the central coefficient and all the
> others. Do you still call it a boxcar?

Regardless of what naming convention is currently causing your
confusion, the following is true:

  -1   -1   -1
  -1    9   -1 
  -1   -1   -1

is identical to  g(x,y) = 10*f(x,y) - f'(x,y) 

where f'(x,y) is f(x,y) convolved with:

 1   1   1
 1   1   1
 1   1   1


> 
> >       But if you apply the smoothing (the box-car part) with many
> > iterations it becomes more gaussian like. And if you give the user
> > control over the amount of that the smoothed part gets subtracted from
> > the original you have an interactive de-blurring tool that could produce
> > results like he is showing.
> 
> He gives no such control his only variable is the number of passes with
> the only kernel he supports.

	I read there were 2 parameters one is number of iterations one is what
he calls depth of focus. this would be analogous to the controls you
would have if you were doing this in a darkroom.

	But I'm not claiming that I am certain that this is what he is doing
only this is one (the easiest to explain) of several possible algorithm
that fits with his vague description of what he is doing . What is
certain is what you are claiming he is doing is incorrect. You can tell
your accusations have been incorrect by just looking at the pictures.  


> 
> >       This is a standard technique for improving blurry images that has been
> > used by photographers since before computers were around. I'm guessing
> > what he is claiming as unique is the method of creating the blurred
> > version -> f'(x,y)
> 
> I must have missed that part. Where does he write about creating a
> blurred version by any method?
> 

see above


>    ...
> 
> >> There is an anomaly in Tontoko's images. Each pass of the filter kernel
> >> leaves garbage at the edges of the image, yet none is visible in the
> >> sharpened images. ...
> 
> No comment on that?

 	The image you were complaining about had a black background. I
wouldn't expect anything at all to happen there. 
	He's offering a free trial of his software. If you are interested in
how he handles the edge conditions you could find out for yourself. The
fact that you can find no evidence that the program behaves as badly as
it would if you had written it, hardly seems like a criticism.

-jim

----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----

jim skrev:
> Jerry Avins wrote:
> >
> > jim wrote:
> > >
> > > Jerry Avins wrote:
> > >
> > >
> > >> That's a sharpening mask, not unsharpening. A typo, I presume?
> > >>
> > >
> > >
> > > No I meant what I said. An unsharp filter is when you subtract a blurred
> > > version from the original:
> > >    g(x,y) = f(x,y) - f'(x,y)
> > >       where f'(x,y) is a blurred version of f(x,y)
> >
> > Yes. I've done that myself. He uses, if the web site is correct, a
> > peaking filter. He claims that a net gain of 1/9 rather than the typical
> > unity is somehow beneficial, but that he doesn't know why.
>
> You are so hung up on your naming conventions for filters you can't see
> the forest for the trees.

Wrong. Both Jerry and I read the posted description of this
processing method, and try and make sense of what is going
on. Our only way to do that, is to try and understand the
terms and information that are posted. The poster uses the
term "Focus Corrector", what is posted on the site is nothing
but a 2D FIR filter. This is a major diuscrepancy in terminology.

> > >       The smoothing filter here is just a box car filter which is a bit
> > > crude.
> >
> > Look at the difference between the central coefficient and all the
> > others. Do you still call it a boxcar?

It is nothing but a 2D FIR filter.

> Regardless of what naming convention is currently causing your
> confusion, the following is true:
>
>   -1   -1   -1
>   -1    9   -1
>   -1   -1   -1
>
> is identical to  g(x,y) = 10*f(x,y) - f'(x,y)
>
> where f'(x,y) is f(x,y) convolved with:
>
>  1   1   1
>  1   1   1
>  1   1   1

I don't see that relation. Could you elaborate?

> > >       But if you apply the smoothing (the box-car part) with many
> > > iterations it becomes more gaussian like. And if you give the user
> > > control over the amount of that the smoothed part gets subtracted from
> > > the original you have an interactive de-blurring tool that could produce
> > > results like he is showing.
> >
> > He gives no such control his only variable is the number of passes with
> > the only kernel he supports.
>
> 	I read there were 2 parameters one is number of iterations one is what
> he calls depth of focus. this would be analogous to the controls you
> would have if you were doing this in a darkroom.

So where does that come into play with this processing? Why isn't
it explained anywhere?

> 	But I'm not claiming that I am certain that this is what he is doing
> only this is one (the easiest to explain) of several possible algorithm
> that fits with his vague description of what he is doing .

It seems everybody agree on two thing: No one knows for sure
what is going on, and the explanation is vague.

>  What is
> certain is what you are claiming he is doing is incorrect.

No one claim he is doung anything. We ask what he does, using the
fact that the explanation he has posted makes little sense.

> You can tell
> your accusations have been incorrect by just looking at the pictures.

Again, no one are accusing anybody for anything. We are asking
questions in order to learn how he does his trick. The posted
description makes little sense.

Rune

Rune Allnor wrote:
 This is a major diuscrepancy in terminology.
> 

Jerry called him a fraud. His description and terminology may be poor,
but that doesn't make him a charlatan. If someone had called him a
miserable failure at explaining how his algorithm works I wouldn't have
disagreed. But despite his apparent incapacity or unwillingness to
explain what he is doing the results do appear to be quite good. 

> > > >       The smoothing filter here is just a box car filter which is a bit
> > > > crude.
> > >
> > > Look at the difference between the central coefficient and all the
> > > others. Do you still call it a boxcar?
> 
> It is nothing but a 2D FIR filter.
> 

Did anybody say it wasn't. But given the hint that there are iterations
involved and some vague mention of variance it's possible it might not
be.
	And no, an unsharp filter is not the same as deconvolution, but it can
(if done right) have very close to the same frequency response with out
the stability issues. Thus it has been for a very long time and still is
used often for correcting focus. Any other method of arriving at  a
frequency response that is approximately the same as deconvolution
should also work well.


> > Regardless of what naming convention is currently causing your
> > confusion, the following is true:
> >
> >   -1   -1   -1
> >   -1    9   -1
> >   -1   -1   -1
> >
> > is identical to  g(x,y) = 10*f(x,y) - f'(x,y)
> >
> > where f'(x,y) is f(x,y) convolved with:
> >
> >  1   1   1
> >  1   1   1
> >  1   1   1
> 
> I don't see that relation. Could you elaborate?


In plain language: 
	Convolving an image with the first filter above produces the same
result as multiplying the image by 10 and then subtracting a version of
the image produced by convolving the image with the second filter. 

	My guess is that the second filter (or some close variant) is what he
is talking about when he mentions iterations.


> 
> > > >       But if you apply the smoothing (the box-car part) with many
> > > > iterations it becomes more gaussian like. And if you give the user
> > > > control over the amount of that the smoothed part gets subtracted from
> > > > the original you have an interactive de-blurring tool that could produce
> > > > results like he is showing.
> > >
> > > He gives no such control his only variable is the number of passes with
> > > the only kernel he supports.
> >
> >       I read there were 2 parameters one is number of iterations one is what
> > he calls depth of focus. this would be analogous to the controls you
> > would have if you were doing this in a darkroom.
> 
> So where does that come into play with this processing? Why isn't
> it explained anywhere?
> 

The explanation is extremely short and vague, but that doesn't justify
declaring the results to be invalid or fraudulent. The OP did not claim
his description of the method was sufficient that anyone could duplicate
his results. He may not even want to give that much information. But
nobody has asked directly for the details. 

-jim



> 
> No one claim he is doung anything. We ask what he does, using the
> fact that the explanation he has posted makes little sense.

I must have missed where you asked for an explanation. You asked me to
explain further you told him his method was unremarkable and the
difference he was seeing was purely due to using floating point
arithmetic versus integer arithmetic.

----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----

jim skrev:
> Rune Allnor wrote:
>  This is a major diuscrepancy in terminology.
> >
>
> Jerry called him a fraud. His description and terminology may be poor,
> but that doesn't make him a charlatan. If someone had called him a
> miserable failure at explaining how his algorithm works I wouldn't have
> disagreed. But despite his apparent incapacity or unwillingness to
> explain what he is doing the results do appear to be quite good.

Without accusing anybody of anything, it might be worth remembering
that one difference between an engineer and a charlatan is that the
engineer is able to both justify and explain how a certain operation
works. That's what happens here every day; questions are asked
and answers are given. One very soon finds out who to trust and
who to be sceptical about, based on the answers they give.

...


> > >       I read there were 2 parameters one is number of iterations one is what
> > > he calls depth of focus. this would be analogous to the controls you
> > > would have if you were doing this in a darkroom.
> >
> > So where does that come into play with this processing? Why isn't
> > it explained anywhere?
> >
>
> The explanation is extremely short and vague, but that doesn't justify
> declaring the results to be invalid or fraudulent.

Again, I just want to get an impression of who is posting these
results.
One have to read the explanation very carefully and dig through one or
two additional links to find the matrix inversions and stability
issues.
The main impression by the first page is that these are mainly
mask operations implemented in floating point arithmetics rather than
fixed point arithmetics.

> The OP did not claim
> his description of the method was sufficient that anyone could duplicate
> his results. He may not even want to give that much information. But
> nobody has asked directly for the details.

Not in this thread. The OP did post a similar example a few
hours earlier:

http://groups.google.no/group/comp.dsp/browse_frm/thread/b0adce1608437c34/#

Maybe I, too, am guilty of not making sufficiently clear statements,
but the sentiment, if not the phrasing, behind my first post was "What
the f**k is this?" The OPs response seem to indicate that he agrees
we are talking about mask operators. Note that he does not correct
my first impression that this is a mask operator; he says "calculation
shows definite difference between them"

> > No one claim he is doung anything. We ask what he does, using the
> > fact that the explanation he has posted makes little sense.
>
> I must have missed where you asked for an explanation. You asked me to
> explain further you told him his method was unremarkable and the
> difference he was seeing was purely due to using floating point
> arithmetic versus integer arithmetic.

There were two similar threads started in a matter of hours, of
which this is the second. The first questions were asked in the
first thread. At least my, and presumably also Jerry's, responses
in this thread are continuations of the discussions started in the
first thread.

Rune

jim wrote:

     ...

> 	I read there were 2 parameters one is number of iterations one is what
> he calls depth of focus. this would be analogous to the controls you
> would have if you were doing this in a darkroom.

I apparently missed that.

> 	But I'm not claiming that I am certain that this is what he is doing
> only this is one (the easiest to explain) of several possible algorithm
> that fits with his vague description of what he is doing . What is
> certain is what you are claiming he is doing is incorrect. You can tell
> your accusations have been incorrect by just looking at the pictures.  

I can see from the pictures that they are sharper, not that they are in 
better focus. All his images are of objects whose actual appearance in 
inaccessible. Wait: I can also see that there are no edge artifacts.

>>>       This is a standard technique for improving blurry images that has been
>>> used by photographers since before computers were around. I'm guessing
>>> what he is claiming as unique is the method of creating the blurred
>>> version -> f'(x,y)
>> I must have missed that part. Where does he write about creating a
>> blurred version by any method?
>>
> 
> see above

I see where you write about it. Where does he?
>>    ...
>>
>>>> There is an anomaly in Tontoko's images. Each pass of the filter kernel
>>>> leaves garbage at the edges of the image, yet none is visible in the
>>>> sharpened images. ...
>> No comment on that?
> 
>  	The image you were complaining about had a black background. I
> wouldn't expect anything at all to happen there. 

Saturn's red spot?

> 	He's offering a free trial of his software. If you are interested in
> how he handles the edge conditions you could find out for yourself. The
> fact that you can find no evidence that the program behaves as badly as
> it would if you had written it, hardly seems like a criticism.

I'll buy that! Did you get the impression that he ever used a mask other 
than

   -1/9   -1/9   -1/9
   -1/9     1    -1/9
   -1/9   -1/9   -1/9
?

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Jerry Avins wrote:
> 

> 
> I'll buy that! Did you get the impression that he ever used a mask other
> than
> 
>    -1/9   -1/9   -1/9
>    -1/9     1    -1/9
>    -1/9   -1/9   -1/9
> ?

	Yes it was obvious. He said he used iterations. If you use that filter
over and over the results would be horrific. So obviously that isn't
what he was saying. A single iteration would produce fair results but
anymore than once, it would get really ugly.

	Even if you missed the part about there being input parameters and
thought he was using a single filter once only you should know it's not
that filter that produced those images. Simply by looking at the
pictures you can tell he is using something better than that. That
filter would leave some blocky artifacts that would be noticeable.
Running that filter over and over would simply magnify the blockiness.
So your interpretation is clearly wrong. According to Rune that makes
you a charlatan because that's what charlatans do - fail to explain how
something works.
	
	I assume he posted here to get people to download and try the program
and offer criticism based on what they find using their own images. I
don't think he was looking for critiques on what people would imagine
they might find.

-jim

----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----

jim wrote:
> 
> Jerry Avins wrote:
> 
>> I'll buy that! Did you get the impression that he ever used a mask other
>> than
>>
>>    -1/9   -1/9   -1/9
>>    -1/9     1    -1/9
>>    -1/9   -1/9   -1/9
>> ?
> 
> 	Yes it was obvious. He said he used iterations. If you use that filter
> over and over the results would be horrific. So obviously that isn't
> what he was saying. A single iteration would produce fair results but
> anymore than once, it would get really ugly.

You're inferring what OP must have done from your knowledge of what 
can't work. I'm trying to understand what he did from what he said he 
did. I agree that what he said he did can't work, and I'l like a 
clarification from him. Tit you catch Rune's observation that this 
thread has only half of his remarks?

> 	Even if you missed the part about there being input parameters and
> thought he was using a single filter once only you should know it's not
> that filter that produced those images. Simply by looking at the
> pictures you can tell he is using something better than that. That
> filter would leave some blocky artifacts that would be noticeable.
> Running that filter over and over would simply magnify the blockiness.
> So your interpretation is clearly wrong. According to Rune that makes
> you a charlatan because that's what charlatans do - fail to explain how
> something works.

Iterating with that filter is equivalent to a single pass with a larger 
filter, it that's what you mean.

> 	I assume he posted here to get people to download and try the program
> and offer criticism based on what they find using their own images. I
> don't think he was looking for critiques on what people would imagine
> they might find.

I'm sure you're right about that. If I get a chance, I'll do that and 
have a go at the three-image comparison I described.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Jerry Avins wrote:

> =

> You're inferring what OP must have done from your knowledge of what
> can't work. =


Well excuse me, I wasn't aware using knowledge was not permitted. I was
also using the information that his images supplied. After all, if "it
is nothing but a 2D FIR filter", as Rune put it, the images themselves
explain everything.

>I'm trying to understand what he did from what he said he
> did.  I agree that what he said he did can't work,

No, you are trying to prove your interpretation of what is written on
the software web page can't work.  =



> and I'l like a clarification from him. =


If you are so desperate to know what the impulse response is for the
processing at each of the different parameter settings available why not
download the program and feed it an impulse and see what the response is
for each setting. =


>Tit you catch Rune's observation that this
> thread has only half of his remarks?

The OP wrote in a different thread:


The pictures shown in the following URL;
http://www.unmannedspaceflight.com/index.php?act=3DAttach&type=3Dpost&id=3D=
2017
are the original image of Vesta (left), its de-convoluted image (middle
/ focus depth:=3D3.8, iteration:=3D8 / software: Focus Corrector ) and
another de-convoluted image (right / focus depth:=3D4.3, iteration:=3D8 /=

software: Focus Corrector).


With his statement "focus depth:=3D4.3, iteration:=3D8" that is enough ri=
ght
there to know he is not applying a simple kernel you can write down with
3 lines of ascii. So I don't know why you continue to beat that dead
horse.

-jim

> =

> >       Even if you missed the part about there being input parameters =
and
> > thought he was using a single filter once only you should know it's n=
ot
> > that filter that produced those images. Simply by looking at the
> > pictures you can tell he is using something better than that. That
> > filter would leave some blocky artifacts that would be noticeable.
> > Running that filter over and over would simply magnify the blockiness=
=2E
> > So your interpretation is clearly wrong. According to Rune that makes=

> > you a charlatan because that's what charlatans do - fail to explain h=
ow
> > something works.
> =

> Iterating with that filter is equivalent to a single pass with a larger=

> filter, it that's what you mean.
> =

> >       I assume he posted here to get people to download and try the p=
rogram
> > and offer criticism based on what they find using their own images. I=

> > don't think he was looking for critiques on what people would imagine=

> > they might find.
> =

> I'm sure you're right about that. If I get a chance, I'll do that and
> have a go at the three-image comparison I described.
> =

> Jerry
> --
> Engineering is the art of making what you want from things you can get.=

> =AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=
=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=
=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF=AF

----== Posted via Newsfeeds.Com - Unlimited-Unrestricted-Secure Usenet News==----
http://www.newsfeeds.com The #1 Newsgroup Service in the World! 120,000+ Newsgroups
----= East and West-Coast Server Farms - Total Privacy via Encryption =----