| |
 |
|
|
Science Forum Index » Image Processing Forum » Image-to-image registration....
Page 1 of 1
|
| Author |
Message |
| Yalin... |
 Posted: Wed May 21, 2008 9:18 am |
|
|
|
Guest
|
Hello:
I am trying to do change detection between to years of a lake, and have had
a hard time getting the two images to line up accurately. I have to make
sure that the accuracy of the two images are <0.5 pixels apart. I was
wondering if anyone can give me some advice or direct me to some
information on the image-to-image registration process. I was using ILWIS
3.4, but now have IDRISI Amdes 15.0. Is it essential that I have GCP for
the two images? |
|
|
| Back to top |
|
| aruzinsky... |
| Posted: Wed May 21, 2008 9:18 am |
|
|
|
Guest
|
On May 21, 8:18 am, Yalin <mjm_v... at (no spam) yahoo.com> wrote:
Quote: Hello:
I am trying to do change detection between to years of a lake, and have had
a hard time getting the two images to line up accurately. I have to make
sure that the accuracy of the two images are <0.5 pixels apart. I was
wondering if anyone can give me some advice or direct me to some
information on the image-to-image registration process. I was using ILWIS
3.4, but now have IDRISI Amdes 15.0. Is it essential that I have GCP for
the two images?
Post a link to the two images.
You cannot do rigid alignment of changed features such as trees that
have grown and/or been cut down. Doing non-rigid alignment would
defeat the purpose of detecting changes. Thus, you should do manual
registration of stable features. To do manual alignment at subpixel
accuracy, it may be necessary to enlarge the images before alignment,
then reduce after alignment.
GCP ?= German Communist Party, http://en.wikipedia.org/wiki/GCP
You should try free demo version of SAR Image Processor,
http://www.general-cathexis.com/ . |
|
|
| Back to top |
|
| Pixel.to.life... |
| Posted: Wed May 21, 2008 9:18 am |
|
|
|
Guest
|
On May 21, 7:18 am, Yalin <mjm_v... at (no spam) yahoo.com> wrote:
Quote: Hello:
I am trying to do change detection between to years of a lake, and have had
a hard time getting the two images to line up accurately. I have to make
sure that the accuracy of the two images are <0.5 pixels apart. I was
wondering if anyone can give me some advice or direct me to some
information on the image-to-image registration process. I was using ILWIS
3.4, but now have IDRISI Amdes 15.0. Is it essential that I have GCP for
the two images?
What kind of images are they? Photographs?
My first guess is you need a rigid transformation first.
Are you willing to write your own code to register?
In medical imaging, we frequently use ITK (Insight ToolKit) for
experimentations. Google for it, you will find it.
You may also want to go over this thread:
http://groups.google.com/group/medicalimagingscience/browse_thread/thread/1ca1f93ab0d9ad97/eaee170781700d3c?lnk=gst&q=registration#eaee170781700d3c |
|
|
| Back to top |
|
| Yalin... |
| Posted: Wed May 21, 2008 7:09 pm |
|
|
|
Guest
|
The images are satellite images (Landsat TM4/5 & ETM+). TM05.TIF is from
2005 and TM93 is from 1993. I have only sent one band, band 3. I am going
to create an NDVI to evaluate the change in biomass of one specific lake,
but I thought resampling the whole image then clipping the the lake out and
creating the NDVI would be best.
I am going do use an easy image differencing technique and then classifing
the changed picks. However, it is crucial that I get the 2005 image to
lineup with the 1993 image.
This is a problem that I have gone over for some time, and I am willing to
do whatever it takes; however, I have no programming experience. This is a
side project that I have decided to take on for a prof. at my University.
The two images are below:
http://download.yousendit.com/D1F9E71666B28C17 <-- TM05
http://download.yousendit.com/9E54964E70312271 <-- TM93
Thank you for the help and guidance! |
|
|
| Back to top |
|
| Yalin... |
| Posted: Wed May 21, 2008 7:12 pm |
|
|
|
Guest
|
Yalin <mjm_vail at (no spam) yahoo.com> wrote in
news:Xns9AA5CD14B3AC8mjmvailyahoocom at (no spam) 127.0.0.1:
Quote: The images are satellite images (Landsat TM4/5 & ETM+). TM05.TIF is
from 2005 and TM93 is from 1993. I have only sent one band, band 3.
I am going to create an NDVI to evaluate the change in biomass of one
specific lake, but I thought resampling the whole image then clipping
the the lake out and creating the NDVI would be best.
I am going do use an easy image differencing technique and then
classifing the changed pixels. However, it is crucial that I get the
2005 image to lineup with the 1993 image.
This is a problem that I have gone over for some time, and I am
willing to do whatever it takes; however, I have no programming
experience. This is a side project that I have decided to take on for
a prof. at my University.
The two images are below:
http://download.yousendit.com/D1F9E71666B28C17 <-- TM05
http://download.yousendit.com/9E54964E70312271 <-- TM93
Thank you for the help and guidance!
|
|
|
| Back to top |
|
| Pixel.to.life... |
| Posted: Wed May 21, 2008 7:23 pm |
|
|
|
Guest
|
On May 21, 9:45 am, aruzinsky <aruzin... at (no spam) general-cathexis.com> wrote:
Quote: On May 21, 8:18 am, Yalin <mjm_v... at (no spam) yahoo.com> wrote:
Hello:
I am trying to do change detection between to years of a lake, and have had
a hard time getting the two images to line up accurately. I have to make
sure that the accuracy of the two images are <0.5 pixels apart. I was
wondering if anyone can give me some advice or direct me to some
information on the image-to-image registration process. I was using ILWIS
3.4, but now have IDRISI Amdes 15.0. Is it essential that I have GCP for
the two images?
Post a link to the two images.
You cannot do rigid alignment of changed features such as trees that
have grown and/or been cut down. Doing non-rigid alignment would
defeat the purpose of detecting changes. Thus, you should do manual
registration of stable features. To do manual alignment at subpixel
accuracy, it may be necessary to enlarge the images before alignment,
then reduce after alignment.
GCP ?= German Communist Party,http://en.wikipedia.org/wiki/GCP
You should try free demo version of SAR Image Processor,http://www.general-cathexis.com/.
Are you talking about locally rigidly registering features like trees?
Trees are an example of features that may have changed (their
projection on the LandSAT may have transformed in a non-rigid manner).
Hence talking about a local rigid registration for non-rigidly
changing features does not make sense in the first place. Why say a
thing that doesnt make sense and then state the obvious?
There could always be rigid structures in the images that do not
change shape. Such features (such as boundary of a lake, or land)
could be registered rigidly (selectively), and the resulting global
transform would give an idea of change in other features. I highly
recommend an automatic affine registration method. Look for opensource
projects (OpenCV e.g.).
If that does not work, you may try a manual affine registration
(rotation, translation, scaling). Unfortunately I am not aware of a
free software that will allow manual registration of LandSat images.
If you google, you will sure find some that are paid software.
And what is all this with finding weird expansions for GCM? If you
google GCM, you dont see this expansion on the first page. Keep the
replies to the point. |
|
|
| Back to top |
|
| Juho Pitkänen... |
| Posted: Thu May 22, 2008 2:31 am |
|
|
|
Guest
|
Quote: The images are satellite images (Landsat TM4/5 & ETM+). TM05.TIF is from
2005 and TM93 is from 1993. I have only sent one band, band 3. I am going
to create an NDVI to evaluate the change in biomass of one specific lake,
but I thought resampling the whole image then clipping the the lake out and
creating the NDVI would be best.
I am going do use an easy image differencing technique and then classifing
the changed picks. However, it is crucial that I get the 2005 image to
lineup with the 1993 image.
This is a problem that I have gone over for some time, and I am willing to
do whatever it takes; however, I have no programming experience. This is a
side project that I have decided to take on for a prof. at my University.
I have sometimes bookmarked a program that looks promising:
http://regima.dpi.inpe.br/
However, I have not tried it myself and their web-demo does not work, at
least with this computer. But there is a standalone program as well,
requiring registration.
If the software works for you, please let us know. |
|
|
| Back to top |
|
| aruzinsky... |
| Posted: Thu May 22, 2008 4:49 am |
|
|
|
Guest
|
On May 21, 6:09 pm, Yalin <mjm_v... at (no spam) yahoo.com> wrote:
Quote: The images are satellite images (Landsat TM4/5 & ETM+). TM05.TIF is from
2005 and TM93 is from 1993. I have only sent one band, band 3. I am going
to create an NDVI to evaluate the change in biomass of one specific lake,
but I thought resampling the whole image then clipping the the lake out and
creating the NDVI would be best.
I am going do use an easy image differencing technique and then classifing
the changed picks. However, it is crucial that I get the 2005 image to
lineup with the 1993 image.
This is a problem that I have gone over for some time, and I am willing to
do whatever it takes; however, I have no programming experience. This is a
side project that I have decided to take on for a prof. at my University.
The two images are below:
http://download.yousendit.com/D1F9E71666B28C17<-- TM05http://download.yousendit.com/9E54964E70312271<-- TM93
Thank you for the help and guidance!
I can't download because I use a dial up modem. Can you convert to
JPEG and repost? |
|
|
| Back to top |
|
| sueb... |
| Posted: Thu May 22, 2008 7:17 am |
|
|
|
Guest
|
On May 21, 5:09 pm, Yalin <mjm_v... at (no spam) yahoo.com> wrote:
Quote: The images are satellite images (Landsat TM4/5 & ETM+). TM05.TIF is from
2005 and TM93 is from 1993. I have only sent one band, band 3. I am going
to create an NDVI to evaluate the change in biomass of one specific lake,
but I thought resampling the whole image then clipping the the lake out and
creating the NDVI would be best.
I am going do use an easy image differencing technique and then classifing
the changed picks. However, it is crucial that I get the 2005 image to
lineup with the 1993 image.
This is a problem that I have gone over for some time, and I am willing to
do whatever it takes; however, I have no programming experience. This is a
side project that I have decided to take on for a prof. at my University.
The two images are below:
http://download.yousendit.com/D1F9E71666B28C17<-- TM05http://download.yousendit.com/9E54964E70312271<-- TM93
Thank you for the help and guidance!
You don't need ground control points but you need to find the same
points on both images. (You would be wise to use GCPs because the
next thing your professor who needs help is going to want to do is
bring in some other spatially-referenced dataset to overlay on
these.) Look for things that are stable like road intersections,
bridge abutments, etc. Get a good spatial distribution of points
around the image. Measure the points as accurately as possible.
One of your images becomes the control and the other is fitted to it.
Your points are used to establish a uniform grid of the desired pixel
size (probably 30m if you're using Landsat TMimages). You fit the
other images points to this grid, analyze with a least squares
criterion, removing points to improve the goodness of fit, and
generate a transformation polynomial describing the fit when your
overall error is lower than one pixel. Then project your second image
into the grid of the first, using the polynomial.
The process is called registering or warping. I'm sure there's
software in Idrisi (which is downloadable?) that would handle this.
Look in any digital image processing text geared to remote sensing
(Jensen) or the Manual of Remote Sensing for a better explanation.
Susan B. |
|
|
| Back to top |
|
| Martin Leese... |
| Posted: Thu May 22, 2008 10:07 am |
|
|
|
Guest
|
aruzinsky wrote:
Quote: On May 21, 8:18 am, Yalin <mjm_v... at (no spam) yahoo.com> wrote:
....
Is it essential that I have GCP for
the two images?
No. GCPs will only help you register the
images to a known map projection. They will
not help you register the two images to each
other.
....
Funny. GCP = Ground Control Point. This is
a feature visible on a satellite image whose
ground coordinates (lat and long) are known.
--
Regards,
Martin Leese
E-mail: please at (no spam) see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/ |
|
|
| Back to top |
|
| Martin Leese... |
| Posted: Thu May 22, 2008 10:16 am |
|
|
|
Guest
|
aruzinsky wrote:
I also use dial-up. The images are over
50 MBytes each. I don't think JPEG versions
are going to help much. I guess we two will
just have crawl into holes and die.
--
Regards,
Martin Leese
E-mail: please at (no spam) see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/ |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Fri Aug 29, 2008 6:31 pm
|
|