| ratio imaging [message #29710] |
Fri, 08 March 2002 08:31  |
Dan Larson
Messages: 21
Registered: March 2002
|
Junior Member |
|
|
I am trying to do simple ratiometric
imaging with IDL. Of course, since
there is some pixelation noise, the
ratio is not very robust. I have
experimented with a number of
different filters (median, Gaussian
deconvolution, smooth) to try and
remove some of this instability.
Is there a filtering technique which
is minimally perturbative that will
remove some numerical artificats
without changing the boundaries of
objects?
Dan Larson
Cornell Univ.
|
|
|
|
| Re: ratio imaging [message #29799 is a reply to message #29710] |
Wed, 13 March 2002 04:39  |
gerhard.holst
Messages: 3
Registered: March 2002
|
Junior Member |
|
|
Dan Larson <drl16@cornell.edu> wrote in message news:<MPG.16f7fce19edad53f989684@newsstand.cit.cornell.edu>...
> In article <9fe17a3c.0203120028.1a7caf67@posting.google.com>,
> gerhard.holst@pco.de says...
>> Craig Markwardt <craigmnet@cow.physics.wisc.edu> wrote in message news:<on1yev3r9q.fsf@cow.physics.wisc.edu>...
>>> Dan Larson <drl16@cornell.edu> writes:
>>>> I am trying to do simple ratiometric
>>>> imaging with IDL. Of course, since
>>>> there is some pixelation noise, the
>>>> ratio is not very robust. I have
>>>> experimented with a number of
>>>> different filters (median, Gaussian
>>>> deconvolution, smooth) to try and
>>>> remove some of this instability.
>>>> Is there a filtering technique which
>>>> is minimally perturbative that will
>>>> remove some numerical artificats
>>>> without changing the boundaries of
>>>> objects?
>>>
>>> Dan, you should be filtering the two images, *before* computing the
>>> ratio, right? I would have said that goes without saying, but now I
>>> am saying it. [ The reason of course is that the ratio does not have
>>> a nice compact statistical distribution, so averaging is less
>>> robust. ]
>>>
>>> Craig
>>
>> Dan,
>> if your are looking for more edge preserving filters you might
>> search for topics like "Savitzky-Golay" and "LOESS", both are
>> filters that might consume a little more time in calculation
>> (especially the LOESS I have found on the web, if you are interested
>> I can look for the link), but they do a good job in smoothing
>> while edge keeping, much better than boxcar, median etc.
>>
>> Gerhard
>>
> Gerhard,
>
> I am familiar with the Savitzky-Golay filter, but I have never used
> the LOESS filter. If you have an implementation that you like, I
> would like to hear about it. Do you know what the acronym stands
> for?
>
> Dan
Hi Dan,
to my knowledge it stands for an old German geological term,
it is no acronym, there is also a filter called Lowess very similar to that.
I have found the origin of the LOESS routines in IDL that I use,
they were developed by H. Freudenreich.
http://www.astro.washington.edu/deutsch/idl/htmlhelp/slibrar y14.html
There is also the source for his routines based on a library from
AT&T Labs. I have attached 2 files (pdf&ps) that give some background
on the filter. If you use the Loess routine of Freudenreich, keep
the pixelwidth odd, otherwise you'll get error messages, if I remember
correct.
Gerhard
|
|
|
|
| Re: ratio imaging [message #29800 is a reply to message #29710] |
Wed, 13 March 2002 03:03 |
the_cacc
Messages: 104
Registered: October 2001
|
Senior Member |
|
|
Yo,
My $0.02... be a bit wary about filtering, it won't do magic for you
(obviously!). However it will do *something*. Be curious about what
that is...
Consider you have N + n1 (numerator + noise) and D + n2 (denominator +
noise) and you are forming (N+n1)/(D+n2). What you really want is N/D,
right? Filtering will give you <N+n1>/<D+n2>, where <> implies some
sort of filtering. Ideally, <N+n1>=N and <D+n2>=D but it won't ever in
reality. If you have an analytical expression for what N and D are,
you might be able to work out what's going on analystically. Otherwise
do some simulations and check that the <N+n1>/<D+n2> you are forming
is approximately the N/D you want. Choose you filter, if any, on this
basis.
Ciao.
|
|
|
|
| Re: ratio imaging [message #29809 is a reply to message #29710] |
Tue, 12 March 2002 08:54 |
Dan Larson
Messages: 21
Registered: March 2002
|
Junior Member |
|
|
In article <9fe17a3c.0203120028.1a7caf67@posting.google.com>,
gerhard.holst@pco.de says...
> Craig Markwardt <craigmnet@cow.physics.wisc.edu> wrote in message news:<on1yev3r9q.fsf@cow.physics.wisc.edu>...
>> Dan Larson <drl16@cornell.edu> writes:
>>> I am trying to do simple ratiometric
>>> imaging with IDL. Of course, since
>>> there is some pixelation noise, the
>>> ratio is not very robust. I have
>>> experimented with a number of
>>> different filters (median, Gaussian
>>> deconvolution, smooth) to try and
>>> remove some of this instability.
>>> Is there a filtering technique which
>>> is minimally perturbative that will
>>> remove some numerical artificats
>>> without changing the boundaries of
>>> objects?
>>
>> Dan, you should be filtering the two images, *before* computing the
>> ratio, right? I would have said that goes without saying, but now I
>> am saying it. [ The reason of course is that the ratio does not have
>> a nice compact statistical distribution, so averaging is less
>> robust. ]
>>
>> Craig
>
> Dan,
> if your are looking for more edge preserving filters you might
> search for topics like "Savitzky-Golay" and "LOESS", both are
> filters that might consume a little more time in calculation
> (especially the LOESS I have found on the web, if you are interested
> I can look for the link), but they do a good job in smoothing
> while edge keeping, much better than boxcar, median etc.
>
> Gerhard
>
Gerhard,
I am familiar with the Savitzky-Golay filter, but I have never used
the LOESS filter. If you have an implementation that you like, I
would like to hear about it. Do you know what the acronym stands
for?
Dan
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
