| Re: Creating a new image from an image input in IDL [message #70796] |
Wed, 05 May 2010 06:57 |
Jeremy Bailin
Messages: 618
Registered: April 2008
|
Senior Member |
|
|
> The only question I have is how would you use a bilinear interpolation
> then. It is important that I take that into account for the after
> picture after the transformation. I am reading in wikipedia what it
> is. I am assuming bilinear interpolation is needed since the
> transformation move pixels to fractional coordinates ??? But in any
> case how would that be done.
Yes that's exactly right, for your problem you will want to do
bilinear interpolation because you'll end up with fractional pixel
coordinates.
The built-in IDL routine INTERPOLATE should work for you. You can use
it by replacing:
oldimage[replicate(chan,npix), oldcoordsX, oldcoordsY]
in the last line with:
interpolate(oldimage[chan,*,*], oldcoordsX, oldcoordsY)
> I am still not clear what this line
> does:
> newcoords = array_indices([nx,ny], lindgen(npix), /dimen)
>
> It initializes an array to lindgen(npix) of the same size of the array
> of original pic???
ARRAY_INDICES maps between 1D indices and 2D coordinates. In other
words, if you have an NX x NY array, you can refer to each element by
its 2D coordinates [X,Y] or by a 1D index X + Y*NX. LINDGEN gives you
a list of all 1D indices, and then ARRAY_INDICES turns those into X,Y
pairs.
> Also for the transformation,
> i.e.,
> oldcoordsX = reform( A * newcoords[0,*] + B * newcoords[1,*], npix)
>
> is oldcoordsX a known or is newcoords a known. From this fragment of
> code, it leads me to believe that newcoords is given and oldcoordsX is
> being solved by the transformation which to me means that oldcoords is
> really the new coordinates after the transformation. Not entirely sure
> if I am understanding your code,but it ran and I saw a before and
> after picture.
Yes, that's correct. You are calculating the coordinates in the old
image (oldcoords) that correspond to a given known position in the new
image (newcoords). If your transformation equation only goes the other
way then things are more complicated...
-Jeremy.
|
|
|
|
| Re: Creating a new image from an image input in IDL [message #70798 is a reply to message #70796] |
Wed, 05 May 2010 06:37  |
bcubeb3
Messages: 3
Registered: May 2010
|
Junior Member |
|
|
On May 5, 8:01 am, Jeremy Bailin <astroco...@gmail.com> wrote:
> On May 5, 2:45 am, bcubeb3 <barry.brian.barr...@gmail.com> wrote:
>
>
>
>> After typing this line of code:
>> IMAGE=READ_TIFF(FILEPATH('/bin/butterfly.tiff'))
>> help, IMAGE
>
>> I get the output
>> IMAGE BYTE = Array[3, 4800, 6000]
>
>> Now I want to write a computer program to systematically loop through
>> each of the n x n pixels of the image and to use a coordinate system
>> in pixel units to compute new coordinates based on the formula theta_s
>> = theta - (size parameter of your choosing in units of
>> pixels)*theta_hat.
>
>> The vector theta_s tells me where to look in the original image to
>> extract intensity information which will then store in my image array.
>> I will use a bilinear scheme when assigning new intensity values that
>> will be stored for my newly created image array theta. Now I have no
>> idea how to even begin. I was looking for stuff online and I was
>> looking at help manuals but all efforts proved futile. Let me know of
>> your suggestions and I greatly appreciate your help on this.
>
>> -Barry
>
> So I think what you're saying (please correct me if I've
> misunderstood!) is that you have a simple transformation between the
> pixel coordinates of the new image and the pixel coordinates of the
> old image. So there's two steps here:
>
> 1) Calculate the transformation for each of the pixel locations of the
> new image
> 2) Copy the image values over
>
> For 1) I would do something like this:
>
> ; size of image:
> imagesize = size(image,/dimen)
> nchan=imagesize[0]
> nx=imagesize[1]
> ny=imagesize[2]
> npix = nx*ny
>
> ; get a 2D list of all pixel coordinates. This is going to take a LOT
> of memory
> ; for a 4800x6000 image!
> newcoords_2d = array_indices([nx,ny], lindgen(npix), /dimen)
>
> ; apply the transformation. I don't really understand your formula,
> ; but hopefully this example will help you.
> ; reform is needed to flatten it instead of being 1xNPIX
> ; X coordinate in old image, as a linear combination of X and Y
> ; coordinates of new image (oldX = A*newX + B*newY)
> oldcoordsX = reform( A * newcoords[0,*] + B * newcoords[1,*], npix)
> ; Y coordinates in new image, as a linear combination of X and Y
> ; coordinates of new image (oldY = C*newX + D*newY)
> oldcoordsY = reform( C * newcoords[0,*] + D * newcoords[1,*], npix)
>
> Once you've done that, copying it over is easy (though again, very
> memory-intensive):
>
> newimage = bytarr(imagesize)
> for chan=0,2 do newimage[chan,*,*] = oldimage[replicate(chan,npix),
> oldcoordsX, oldcoordsY]
>
> (NOTE: COMPLETELY UNTESTED)
>
> -Jeremy.
THIS IS RIGHT you understood my problem. I ran your code and it
works.
The only question I have is how would you use a bilinear interpolation
then. It is important that I take that into account for the after
picture after the transformation. I am reading in wikipedia what it
is. I am assuming bilinear interpolation is needed since the
transformation move pixels to fractional coordinates ??? But in any
case how would that be done. I can tweak the transformation, but I
think you gave me a great start. I am new so it took a while to
understand what your code did. I am still not clear what this line
does:
newcoords = array_indices([nx,ny], lindgen(npix), /dimen)
It initializes an array to lindgen(npix) of the same size of the array
of original pic???
Also for the transformation,
i.e.,
oldcoordsX = reform( A * newcoords[0,*] + B * newcoords[1,*], npix)
is oldcoordsX a known or is newcoords a known. From this fragment of
code, it leads me to believe that newcoords is given and oldcoordsX is
being solved by the transformation which to me means that oldcoords is
really the new coordinates after the transformation. Not entirely sure
if I am understanding your code,but it ran and I saw a before and
after picture.
best,
Barry
|
|
|
|
| Re: Creating a new image from an image input in IDL [message #70802 is a reply to message #70798] |
Wed, 05 May 2010 05:01 |
Jeremy Bailin
Messages: 618
Registered: April 2008
|
Senior Member |
|
|
On May 5, 2:45 am, bcubeb3 <barry.brian.barr...@gmail.com> wrote:
> After typing this line of code:
> IMAGE=READ_TIFF(FILEPATH('/bin/butterfly.tiff'))
> help, IMAGE
>
> I get the output
> IMAGE BYTE = Array[3, 4800, 6000]
>
> Now I want to write a computer program to systematically loop through
> each of the n x n pixels of the image and to use a coordinate system
> in pixel units to compute new coordinates based on the formula theta_s
> = theta - (size parameter of your choosing in units of
> pixels)*theta_hat.
>
> The vector theta_s tells me where to look in the original image to
> extract intensity information which will then store in my image array.
> I will use a bilinear scheme when assigning new intensity values that
> will be stored for my newly created image array theta. Now I have no
> idea how to even begin. I was looking for stuff online and I was
> looking at help manuals but all efforts proved futile. Let me know of
> your suggestions and I greatly appreciate your help on this.
>
> -Barry
So I think what you're saying (please correct me if I've
misunderstood!) is that you have a simple transformation between the
pixel coordinates of the new image and the pixel coordinates of the
old image. So there's two steps here:
1) Calculate the transformation for each of the pixel locations of the
new image
2) Copy the image values over
For 1) I would do something like this:
; size of image:
imagesize = size(image,/dimen)
nchan=imagesize[0]
nx=imagesize[1]
ny=imagesize[2]
npix = nx*ny
; get a 2D list of all pixel coordinates. This is going to take a LOT
of memory
; for a 4800x6000 image!
newcoords_2d = array_indices([nx,ny], lindgen(npix), /dimen)
; apply the transformation. I don't really understand your formula,
; but hopefully this example will help you.
; reform is needed to flatten it instead of being 1xNPIX
; X coordinate in old image, as a linear combination of X and Y
; coordinates of new image (oldX = A*newX + B*newY)
oldcoordsX = reform( A * newcoords[0,*] + B * newcoords[1,*], npix)
; Y coordinates in new image, as a linear combination of X and Y
; coordinates of new image (oldY = C*newX + D*newY)
oldcoordsY = reform( C * newcoords[0,*] + D * newcoords[1,*], npix)
Once you've done that, copying it over is easy (though again, very
memory-intensive):
newimage = bytarr(imagesize)
for chan=0,2 do newimage[chan,*,*] = oldimage[replicate(chan,npix),
oldcoordsX, oldcoordsY]
(NOTE: COMPLETELY UNTESTED)
-Jeremy.
|
|
|
|
| Re: Creating a new image from an image input in IDL [message #70804 is a reply to message #70802] |
Wed, 05 May 2010 04:38 |
bcubeb3
Messages: 3
Registered: May 2010
|
Junior Member |
|
|
On May 5, 2:45 am, bcubeb3 <barry.brian.barr...@gmail.com> wrote:
> After typing this line of code:
> IMAGE=READ_TIFF(FILEPATH('/bin/butterfly.tiff'))
> help, IMAGE
>
> I get the output
> IMAGE BYTE = Array[3, 4800, 6000]
>
> Now I want to write a computer program to systematically loop through
> each of the n x n pixels of the image and to use a coordinate system
> in pixel units to compute new coordinates based on the formula theta_s
> = theta - (size parameter of your choosing in units of
> pixels)*theta_hat.
>
> The vector theta_s tells me where to look in the original image to
> extract intensity information which will then store in my image array.
> I will use a bilinear scheme when assigning new intensity values that
> will be stored for my newly created image array theta. Now I have no
> idea how to even begin. I was looking for stuff online and I was
> looking at help manuals but all efforts proved futile. Let me know of
> your suggestions and I greatly appreciate your help on this.
>
> -Barry
FOllow up.
Sorry for not being specific. The following describes the algorithm I
have to code up but can't figure how to set up a blank n x n pixel
image or how to compute new coordinates theta using a transformation
equation of the following form: θ⃗s = θ⃗ − θE θˆ, where θE is any size
of your choosing or how to use the bilinear interpolation scheme to
extract the intensity for each pixel. This is a gravitational lensing
problem. Any help of any form is greatly appreciated. I am a newbie so
I am new to IDL and it's hard for me to find the routine that does the
trick since there are so many of them and I tried google with no luck.
You don't have to solve the problem just a direction of where to start
is much appreciated.
• Choose a size (in units of pixels) for the Einstein radius.
• Set up a blank n × n-pixel image-plane array (to match in size your
n × n source image).
• Set up your computer program to systematically loop through each of
the n × n pixels in your image-plane array.
• For each pixel, use the θx,θy coordinate (in dimensionless pixel
units) to compute θ⃗ and θˆ, to compute θ⃗s.
• The vector θ⃗s tells you where to look in the original source image
to extract the intensity which you will then store in your image array
at location θ⃗
• In general, however, θ⃗s will not point to the center of one of the
pixels in the original image. Therefore, you will need to interpolate
the intensity among the nearest four pixels. A bilinear interpolation
scheme is outlined below.
• Finally, after looping through all pixels in the image plane, and
assigning the appropriate intensity values, you will have constructed
the lensed image.
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
