| Re: Finding the mean of a set of images [message #32578] |
Tue, 22 October 2002 10:48  |
Jaco van Gorkom
Messages: 97
Registered: November 2000
|
Member |
|
|
"David Oesch" <oesch@giub.unibe.ch> wrote in message news:3DB56D82.2090407@giub.unibe.ch...
> Craig Markwardt wrote:
>> David Oesch <oesch@giub.unibe.ch> writes:
>>> ...
>>> Does anyone have an algorithm for finding the mean/standardeviation etc
>>> at each pixel position for a set of equal size 2-D images? Currently the
>>> only way I have to do this is to extract all the values for a given
>>> pixel position into a 1-D array and find the mean/standardeviation etc
>>> on that. Doing it pixel by pixel like this is inefficient in IDL so I am
>>> looking for an *array* based algorithm that would find all
>>> the mean/standardeviation etc in parallel.
>>
>> Sure, if you stack your image into a 3D image cube, then you would
>> have something like IMAGE = FLTARR(NX, NY, NIMAGES)
>>
>> Then the mean image is:
>>
>> mean = total(image,3)/nimages
>>
>> ...
> Yes, I got that one with TOTAL, the problem is, it only works with
> imagesets where all datalayers consists of 100% valid pixels...but this
> is not always the case...
> to eliminate those values, I came as far as
>
> mean = total((image NE -9999),3)/nimages
>
> but the problem is, nimages needs to be changed to...
Yes. I suppose you mean total( image * (image NE -9999),3....
How about
valid = image NE -9999
mean = TOTAL(image * valid, 3) / TOTAL(valid, 3)
Or, if you would use NaN (Not-a-Number, floating point!) as missing value:
mean = ( SMOOTH(image, [1,1,Nimages], /NAN) )[*, *, Nimages/2]
For that to work you need to enforce an odd number of images.
cheers,
jaco
|
|
|
|
| Re: Finding the mean of a set of images [message #32582 is a reply to message #32578] |
Tue, 22 October 2002 09:33  |
Jaco van Gorkom
Messages: 97
Registered: November 2000
|
Member |
|
|
"Craig Markwardt" <craigmnet@cow.physics.wisc.edu> wrote in message
news:onznt6k38x.fsf@cow.physics.wisc.edu...
> David Oesch <oesch@giub.unibe.ch> writes:
>> ...
>> Does anyone have an algorithm for finding the mean/standardeviation etc
>> at each pixel position for a set of equal size 2-D images? Currently the
>> only way I have to do this is to extract all the values for a given
>> pixel position into a 1-D array and find the mean/standardeviation etc
>> on that. Doing it pixel by pixel like this is inefficient in IDL so I am
>> looking for an *array* based algorithm that would find all
>> the mean/standardeviation etc in parallel. ...
>
> Sure, if you stack your image into a 3D image cube, then you would
> have something like IMAGE = FLTARR(NX, NY, NIMAGES)
>
> Then the mean image is:
>
> mean = total(image,3)/nimages
>
> The standard deviation is:
>
> meancube = rebin(reform(mean,nx,ny,1),nx,ny,nimages)
> std = sqrt(total((image - meancube)^2,3)/(nimages-1))
>
> Now, what you meant by "etc" can get a little hairier. If you want to
> do median you are probably in trouble, but min and max are easy too:
>
> minimage = image(*,*,0)
> maximage = minimage
> for i = 1, nimages-1 do begin
> minimage = minimage < image(*,*,i)
> maximage = maximage > image(*,*,i)
> endfor
I believe that IDL 5.5 offers the luxury of
maximage = MAX(image, MIN=minimage, DIMENSION=3)
As for median, well, there was a thread on "Finding the median of a set
of images" back in '96. Should be valid still, I guess. Improvising with
transpose(), reform(), median(image, Nimages) and rebin() should be fun, but
not be very fast.
Jaco
|
|
|
|
| Re: Finding the mean of a set of images [message #32583 is a reply to message #32582] |
Tue, 22 October 2002 08:23 |
David Oesch
Messages: 31
Registered: October 2000
|
Member |
|
|
Craig Markwardt wrote:
> David Oesch <oesch@giub.unibe.ch> writes:
>
>> Hello outthere,
>>
>> I know this topic was up before, but all I could find in the list was to
>> go for CALL_EXTERNAL and use a FORTRAN etc. program. Here's my problem:
>> Does anyone have an algorithm for finding the mean/standardeviation etc
>> at each pixel position for a set of equal size 2-D images? Currently the
>> only way I have to do this is to extract all the values for a given
>> pixel position into a 1-D array and find the mean/standardeviation etc
>> on that. Doing it pixel by pixel like this is inefficient in IDL so I am
>> looking for an *array* based algorithm that would find all
>> the mean/standardeviation etc in parallel.
>> Any progs so far in IDL for this problem?..or a decent fortran or C program?
>>
>
>
> Sure, if you stack your image into a 3D image cube, then you would
> have something like IMAGE = FLTARR(NX, NY, NIMAGES)
>
> Then the mean image is:
>
> mean = total(image,3)/nimages
>
> The standard deviation is:
>
> meancube = rebin(reform(mean,nx,ny,1),nx,ny,nimages)
> std = sqrt(total((image - meancube)^2,3)/(nimages-1))
>
> Now, what you meant by "etc" can get a little hairier. If you want to
> do median you are probably in trouble, but min and max are easy too:
>
> minimage = image(*,*,0)
> maximage = minimage
> for i = 1, nimages-1 do begin
> minimage = minimage < image(*,*,i)
> maximage = maximage > image(*,*,i)
> endfor
>
> It's a loop, but unless you have a bazillion images, it will be fast.
>
> Craig
>
Yes, I got that one with TOTAL, the problem is, it only works with
imagesets where all datalayers consists of 100% valid pixels...but this
is not always the case...
to eliminate those values, I came as far as
mean = total((image NE -9999),3)/nimages
but the problem is, nimages needs to be changed to...
up to now, I use the following simple function to get to my data...but
it is time consuming ..
function mean3d,file,nvalid=nvalid
;*********************************************************** ****************
; finding the mean at each pixel position for a set of equal size 2-D images
;*********************************************************** ****************
;file= 3d array
;nvalid= novalid values
;calling sequence: a=mean3d(array,nvalid=-9999)
s=SIZE(file)
outarray=FLTARR(s(1),s(2))
i=0l
FOR i=0,fix(s(1))-1 DO BEGIN
j=0l
FOR j=0,fix(s(2))-1 DO BEGIN
check=WHERE(file(i,j,*) NE nvalid,count_check)
IF count_check GT 0 THEN BEGIN
outarray(i,j)=MEAN(file(i,j,check))
ENDIF
ENDFOR
ENDFOR
RETURN,outarray
END
|
|
|
|
| Re: Finding the mean of a set of images [message #32587 is a reply to message #32583] |
Tue, 22 October 2002 06:33 |
Craig Markwardt
Messages: 1869
Registered: November 1996
|
Senior Member |
|
|
David Oesch <oesch@giub.unibe.ch> writes:
> Hello outthere,
>
> I know this topic was up before, but all I could find in the list was to
> go for CALL_EXTERNAL and use a FORTRAN etc. program. Here's my problem:
> Does anyone have an algorithm for finding the mean/standardeviation etc
> at each pixel position for a set of equal size 2-D images? Currently the
> only way I have to do this is to extract all the values for a given
> pixel position into a 1-D array and find the mean/standardeviation etc
> on that. Doing it pixel by pixel like this is inefficient in IDL so I am
> looking for an *array* based algorithm that would find all
> the mean/standardeviation etc in parallel.
> Any progs so far in IDL for this problem?..or a decent fortran or C program?
Sure, if you stack your image into a 3D image cube, then you would
have something like IMAGE = FLTARR(NX, NY, NIMAGES)
Then the mean image is:
mean = total(image,3)/nimages
The standard deviation is:
meancube = rebin(reform(mean,nx,ny,1),nx,ny,nimages)
std = sqrt(total((image - meancube)^2,3)/(nimages-1))
Now, what you meant by "etc" can get a little hairier. If you want to
do median you are probably in trouble, but min and max are easy too:
minimage = image(*,*,0)
maximage = minimage
for i = 1, nimages-1 do begin
minimage = minimage < image(*,*,i)
maximage = maximage > image(*,*,i)
endfor
It's a loop, but unless you have a bazillion images, it will be fast.
Craig
--
------------------------------------------------------------ --------------
Craig B. Markwardt, Ph.D. EMAIL: craigmnet@cow.physics.wisc.edu
Astrophysics, IDL, Finance, Derivatives | Remove "net" for better response
------------------------------------------------------------ --------------
|
|
|
|
| Re: Finding the mean of a set of images [message #32662 is a reply to message #32582] |
Wed, 23 October 2002 07:28 |
David Oesch
Messages: 31
Registered: October 2000
|
Member |
|
|
<html>
<head>
</head>
<body>
<br>
<br>
Jaco van Gorkom wrote:<br>
<blockquote type="cite" cite="mid:ap3uhr$erme$1@zam602.zam.kfa-juelich.de">
<pre wrap="">"Craig Markwardt" <a class="moz-txt-link-rfc2396E" href="mailto:craigmnet@cow.physics.wisc.edu"><craigmnet@cow.physics.wisc.edu></a> wrote in message<br><a class="moz-txt-link-freetext" href="news:onznt6k38x.fsf@cow.physics.wisc.edu">news:onznt6k38x.fsf@cow.physics.wisc.edu</a>...<br></pre>
<blockquote type="cite">
<pre wrap="">David Oesch <a class="moz-txt-link-rfc2396E" href="mailto:oesch@giub.unibe.ch"><oesch@giub.unibe.ch></a> writes:<br></pre>
<blockquote type="cite">
<pre wrap=""> ...<br>Does anyone have an algorithm for finding the mean/standardeviation etc<br>at each pixel position for a set of equal size 2-D images? Currently the<br>only way I have to do this is to extract all the values for a given<br>pixel position into a 1-D array and find the mean/standardeviation etc<br>on that. Doing it pixel by pixel like this is inefficient in IDL so I am<br>looking for an *array* based algorithm that would find all<br>the mean/standardeviation etc in parallel. ...<br></pre>
</blockquote>
<pre wrap="">Sure, if you stack your image into a 3D image cube, then you would<br>have something like IMAGE = FLTARR(NX, NY, NIMAGES)<br><br>Then the mean image is:<br><br> mean = total(image,3)/nimages<br><br>The standard deviation is:<br><br> meancube = rebin(reform(mean,nx,ny,1),nx,ny,nimages)<br> std = sqrt(total((image - meancube)^2,3)/(nimages-1))<br><br>Now, what you meant by "etc" can get a little hairier. If you want to<br>do median you are probably in trouble, but min and max are easy too:<br><br> minimage = image(*,*,0)<br> maximage = minimage<br> for i = 1, nimages-1 do begin<br> minimage = minimage < image(*,*,i)<br> maximage = maximage > image(*,*,i)<br> endfor<br></pre>
</blockquote>
<pre wrap=""><!----><br>I believe that IDL 5.5 offers the luxury of<br> maximage = MAX(image, MIN=minimage, DIMENSION=3)<br><br>As for median, well, there was a thread on "Finding the median of a set<br>of images" back in '96. Should be valid still, I guess. Improvising with<br>transpose(), reform(), median(image, Nimages) and rebin() should be fun, but<br>not be very fast.<br><br> Jaco<br><br><br></pre>
</blockquote>
<pre class="moz-signature" cols="$mailwrapcol">Well it works fine in 5.5 , but this feature is not explained in the help of the idl 5.5 WIN/PC edition ... in 5.5 the help is obviously still the help of 5.4... </pre>
<br>
</body>
</html>
|
|
|
|
| Re: Finding the mean of a set of images [message #32667 is a reply to message #32582] |
Tue, 22 October 2002 22:18 |
David Oesch
Messages: 31
Registered: October 2000
|
Member |
|
|
<html>
<head>
</head>
<body>
<br>
<br>
Jaco van Gorkom wrote:<br>
<blockquote type="cite" cite="mid:ap3uhr$erme$1@zam602.zam.kfa-juelich.de">
<pre wrap="">"Craig Markwardt" <a class="moz-txt-link-rfc2396E" href="mailto:craigmnet@cow.physics.wisc.edu"><craigmnet@cow.physics.wisc.edu></a> wrote in message<br><a class="moz-txt-link-freetext" href="news:onznt6k38x.fsf@cow.physics.wisc.edu">news:onznt6k38x.fsf@cow.physics.wisc.edu</a>...<br></pre>
<blockquote type="cite">
<pre wrap="">David Oesch <a class="moz-txt-link-rfc2396E" href="mailto:oesch@giub.unibe.ch"><oesch@giub.unibe.ch></a> writes:<br></pre>
<blockquote type="cite">
<pre wrap=""> ...<br>Does anyone have an algorithm for finding the mean/standardeviation etc<br>at each pixel position for a set of equal size 2-D images? Currently the<br>only way I have to do this is to extract all the values for a given<br>pixel position into a 1-D array and find the mean/standardeviation etc<br>on that. Doing it pixel by pixel like this is inefficient in IDL so I am<br>looking for an *array* based algorithm that would find all<br>the mean/standardeviation etc in parallel. ...<br></pre>
</blockquote>
<pre wrap="">Sure, if you stack your image into a 3D image cube, then you would<br>have something like IMAGE = FLTARR(NX, NY, NIMAGES)<br><br>Then the mean image is:<br><br> mean = total(image,3)/nimages<br><br>The standard deviation is:<br><br> meancube = rebin(reform(mean,nx,ny,1),nx,ny,nimages)<br> std = sqrt(total((image - meancube)^2,3)/(nimages-1))<br><br>Now, what you meant by "etc" can get a little hairier. If you want to<br>do median you are probably in trouble, but min and max are easy too:<br><br> minimage = image(*,*,0)<br> maximage = minimage<br> for i = 1, nimages-1 do begin<br> minimage = minimage < image(*,*,i)<br> maximage = maximage > image(*,*,i)<br> endfor<br></pre>
</blockquote>
<pre wrap=""><!----><br>I believe that IDL 5.5 offers the luxury of<br> maximage = MAX(image, MIN=minimage, DIMENSION=3)<br><br>As for median, well, there was a thread on "Finding the median of a set<br>of images" back in '96. Should be valid still, I guess. Improvising with<br>transpose(), reform(), median(image, Nimages) and rebin() should be fun, but<br>not be very fast.<br><br> Jaco<br><br><br></pre>
</blockquote>
Ok, I think now is the time to upgrade from IDL 5.4 to 5.5 :-)....<br>
<br>
Thanks for your tips<br>
<br>
Dave<br>
</body>
</html>
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
