This lab contained exercises that practiced the image preprocessing exercise, geometric correction. Two different types of geometric correction are practiced in this lab. These two major types of geometric preprocessing are usually performed on satellite images prior to the extraction of biophysical and sociocultural information from the satellite images.
For Part 1, in this lab I used a United States Geological Survey (USGS) 7.5 minute digital raster graphic (DRG) image of the Chicago Metropolitan Statistical Area and bordering areas to correct a Landsat TM image covering the same location. In order to perform the geometric correction I had to collect ground control points from the USGS 7.5 minutes DRG to fix the TM image. In Part 2, I used an already corrected Landsat TM image to adjust a geometrically distorted image of the same location.
Methods:
Part 1:
I started by opening an Erdas Imagine viewer and placing the viewer in a display two views setting so that two viewer split the screen. In the first viewer I opened up the image Chicago_drg.img. This image is the USGS 7.5 minute digital raster graphic covering a section of the Chicago region. In the second viewer I opened the image Chicago_2000.img, a Landsat TM image of the same area as the DRG. I made the first viewer holding the DRG image active. I then selected the Thematic heading on the main toolbar to display more options, from which I selected the Control Points option. The Set Geometric Model Window opened up. Under the Select Geometric Model heading I scrolled down and checked the Polynomial option and clicked OK. I accepted the default settings for the Image Layer (New Viewer) and clicked OK. In the next screen I added the Landsat TM image of Chicago and clicked OK. I accepted the Polynomial Model Properties by clicking close.
The Multipoint Geometric Correction window is displayed and I set it to full screen to maximize the viewer. The viewer contained a split screen view of the DRG image in one and the Landsat TM image in the other. The bottom portion of the viewer contained a list of all of the ground control points logged. I made sure there were no ground control points already being used and if there were I deleted them in the bottom portion of the screen where they would be listed. I right clicked each individual entry and selected Delete Selection. I then added four pairs of ground control points between the two images. This image pair only requires three because it is a first order polynomial, but it is always better to collect more than the minimum required ground control points. I did this by aligning the two images to the same area and finding the exact same feature in both images. I then selected the Create GCP tool from the top portion of the Multipoint Geometric Correction interface. I added a ground control point on the same location in both viewers. I changed the color of the point to purple to make it earlier to see. I did this by using the drop down menu in the lower half of the screen under the heading Color. I repeated these steps three more times to create my four ground control points. After all of my points were placed I had to make sure they were in accurate position by checking my Root Mean Square error (RMS). This value is found in the lower right portion of the Multipoint Geometric Correction window. For this part of the lab, the goal was to obtain a RMS error below 2.0. The ideal RMS error is 0.5 and below. To lower the RMS error I zoomed into the ground control points and clicked and held down the cursor and moved the mouse to move the point slightly. I did this while monitoring the total RMS count and found a good spot where the error had dropped. I did this for all of the points until I reached the desired RMS error. I was able to achieve a 0.8752 RMS error.
Once I achieved my RMS error less than 2.0 I selected the Display Resample Image Dialog button on the Multipoint Correction toolbar to create my new image. I set the location and named my output image. I left all of the parameters in the resample image window at their default values and clicked OK to run the tool. At the end of the run I clicked dismiss and close. I exited out of the Multipoint Geometric Correction window and selected not to save the current model. And continued selecting No in the popups that ask if I am sure I do not want to save. I went to the location where I had saved my output image and brought it into the first viewer to see the changes that were done to the new image.
Part 2:
I began by opening a new viewer in Erdas Imagine. I displayed the image sierra_leone_east1991.img in the viewer. This image contains much geometric distortion and to be able to identify this distortion I overlaid the reference image s1_reference_image.img in the same viewer. To visualize the distortion more I activated the swipe function by right clicking on the images on the viewer and clicking swipe. A Viewer Swipe dialog appeared on the main toolbar and I used the slider tab and moved it back and forth across the screen to move the top image out of the way as the slider was moved to notice the differenced between the two images. The swipe function allows me to fully evaluate the level of distortion.
Next, I closed the Viewer Swipe window and cleared the images from the viewer. I set the viewer to the display two views setting and placed the distorted image (sierra_leone_east1991.img) in the first viewer and the reference image (s1_reference_image.img) in the second viewer. I made the viewer containing the distorted image the active viewer. I then selected the Multispectral heading on the main toolbar to display more options, from which I selected the Control Points option. The Set Geometric Model Window opened up. Under the Select Geometric Model heading I scrolled down and checked the Polynomial option and clicked OK. I clicked OK on the Collect Reference Points Form in the GCP Tool Reference Setup. In the next screen I added the reference image and clicked OK on the Reference Map Information. In the Polynomial Model Properties dialog I changed the polynomial order to 3 and clicked Close.
The Multipoint Geometric Correction window is displayed and I set it to full screen to maximize the viewer. The viewer contained a split screen view of the distorted image in one and the reference image in the other. The bottom portion of the viewer contained a list of all of the ground control points logged. I made sure there were no ground control points already being used and if there were I deleted them in the bottom portion of the screen where they would be listed. I right clicked each individual entry and selected Delete Selection. I then added twelve pairs of ground control points between the two images. This image pair only requires ten because it is a third order polynomial, but it is always better to collect more than the minimum required ground control points. I did this by aligning the two images to the same area and finding the exact same feature in both images. I then selected the Create GCP tool from the top portion of the Multipoint Geometric Correction interface. I added a ground control point on the same location in both viewers. I repeated these steps eleven more times to create my four ground control points. After all of my points were placed I had to make sure they were in accurate position by checking my Root Mean Square error (RMS). This value is found in the lower right portion of the Multipoint Geometric Correction window. For this part of the lab, the goal was to obtain a RMS error below 1.0. The ideal RMS error is 0.5 and below. To lower the RMS error I zoomed into the ground control points and clicked and held down the cursor and moved the mouse to move the point slightly. I did this while monitoring the total RMS count and found a good spot where the error had dropped. I did this for all of the points until I reached the desired RMS error. I was able to achieve a 0.8347 RMS error.
Once I achieved my RMS error less than 1.0 I selected the Display Resample Image Dialog button on the Multipoint Correction toolbar to create my new image. I set the location and named my output image. I changed the resample method to bilinear interpolation and left all of the other parameters in the resample image window at their default values and clicked OK to run the tool. At the end of the run I clicked dismiss and close. I exited out of the Multipoint Geometric Correction window and selected not to save the current model. And continued selecting No in the popups that ask if I am sure I do not want to save. I went to the location where I had saved my output image and brought it into the second viewer and overlaid the corrected image over the reference image to see the changes that were done to the new image.
Results:
Part 1: Image-to-map rectification
Figure 1
Figure 1 shows a screen capture of the two image display of the DRG image and Landsat TM image before the rectification. On the left is the image Chicago_drg.img. This image is the USGS 7.5 minute digital raster graphic covering a section of the Chicago region. On the right is the image Chicago_2000.img, a Landsat TM image of the same area as the DRG.
Figure 2
Figure 2 shows a screen capture of the Multipoint Geometric Correction Window after all of my ground control points were entered in. The points were adjusted to create a total RMS error of 0.8752. You can find the total RMS error count in the bottom right corner of this image.
Figure 3
Figure 3 shows a screen capture of an Erdas Imagine viewer in split view with the original Landsat TM image on the left and the newly geometric corrected image. Looking at the two images with the bare eye few changes can be detected, but there has been changes made to the new image.
Part 2: Image to image rectification
Figure 4
Figure 4 shows the geometric distorted image sierra_leone_east1991.img with the reference image s1_reference_image.img overlaid in the same viewer. The images do not overlay on top of each other, so distortion is present.
Figure 5
Figure 5 shows a screen capture of the Multipoint Geometric Correction Window after all of my ground control points were entered in. The points were adjusted to create a total RMS error of 0.8347. You can find the total RMS error count in the bottom right corner of this image. In this figure you can also see where all twelve of the ground control points were placed between the two images.
Figure 6
Figure 6 shows the reference image s1_reference_image.img with the newly geometric corrected image overlaid in the same viewer. The image shows that the new image has better geometric coordinates and is more aligned with the reference image.
Sources:
Satellite images are from Earth Resources Observation and Science Center, United States
Geological Survey.Digital raster graphic (DRG) is from Illinois Geospatial Data Clearing House.
