Lab 6

Introduction

The goal of this lab was to introduce students a image pre-processing tool called geometric correction.  The lab was designed to develop skills on the two major types of geometric correction, image-to-map rectification and image to image registration.

Methods

Part 1

In part one a United States Geological Survey (USGS) 7.5 minute digital raster graphic image of the Chicago Metropolitan Statistical Area was used to correct a Landsat TM image of the same Chicago area.  This was done by collecting ground control points from the USGS image to rectify the TM image.  

In ERDAS IMAGINE, the images Chicago_drg and Chicago_2000 were opened in two separate viewers and fit to frame.  With the image Chicage_2000 highlighted, next Multisprecrtal>Control Points were hit to open up a window titled, Set Geometric Model.  Under "Select Geometric Model" Polynomial was chosen and ok was clicked. 

Figure 1: GCP Tool Reference Setup

Next the window seen if figure 1 above and the default was accepted and OK was clicked.  Then I navigated to the folder containing, Chicago_drg, and it was added. OK was clicked on the reference Map Information dialog.  This will bring up the Polynomial Model Properties window and the default settings were accepted by clicking close.  Now in the Multipoint Geometric Correction window the two Chicago images appeared, figure 2, and the geometric correction was performed.

Figure 2: Multipoint Geometric Correction window containing the two
Chicago images. 

First the GCP's were deleted to to start the process at its first step.  This was done by holding down the shift key and clicking each of the GCPs. Once they were all selected a right clicked was performed and delete selection was clicked.  This should delete all the GCPs except one with no X or Y coordinates.

Both images were right clicked and fit image to window was chosen to correctly fit the image in the frame.  Now the create GCP too, figure 3, was chosen to create GCP points.  

Figure 3: Create GCP tool.

A total of four ground control points were added to image in the same spot in each image to rectify the image.  Each time the GCP buttom was clicked to at a GCP in the same spot to each image looking like figure 4 below.  Under the control center the color purple was chosen simply by clicking the empty space for each GCP. 

Figure 4: The two images with 4 GCP's each in the same location to rectify the image. 

An important component of placing the GCPs in the same position in each image is the RMS error.  Total RMS error can be found in the lower right of the screen and for part one an RMS error under 2 is required.  To reduce RMS error I zoomed way in to each GCP and choose a landmark or coastline to match each GCP with.  It took a while to play around with and obtain a total RMS error under 2 but it was done.  

Figure 5: example showing total RMS error. 

Once the RMS error was under 2 the display re sample image dialog button was chosen and a new name was chosen for the output image.  The function was run and brought into ERDAS IMAGINE.  The results can be seen in the results section below.

Part 2

The directions as part one except this time the correction will be image to image with Sierra Leone images.. 

To do a image to image function the same directions were followed from part 1 except a few changes.  This time 12 GCPs were choose because polynomial order 3 was performed.  While adding 12 GCP's throughout the map a RMS error under 1 was desired.  Comparing this to part 1 it was much tougher because the RMS was lower and their was 12 points.

Figure 6: The 12 GCPs were set in place similar to the image above. 

After the 12 GCPs were added and a RMS below 1 was achieved the display re sample image dialog button was clicked and an output name was chosen.  In the re sample method option it was changed from nearest neighbor to bilinear interpolation because it called for better spatial accuracy.  The model was run and the output image was brought into ERDAS IMAGINE for results. 

Results

Part 1

Figure 7: Rectified Chicago area image. 

Figure 8: Total RMS error of 1.54

I found the image to map conversion to be more difficult and not as clean as the image to image correction.  The map made it really difficult to find the correct location of GCPs to lower the RMS error.  Figure 9 below was very accurate in terms of features and elements in the image.  Image to map corrects more of a city landscape I believe compared to image to image correcting a landscape.


Part 2

Figure 9: Image to Image registration of Sierra Leone

Figure 10: Total RMS error of .2514

How geometrically correct is your rectified image? In other words, how spatially accurate is 

it in relation to the reference image you used? 

    The new image is very geometrically correct. Even though there is a lot of cloud cover compared to the old images when zooming in on the new image land forms and shapes are much clearer and easier to interpret compared to the old images.  The old image was much distorted and after correcting it the landforms are much clearer because they were fixed to be connected together in the correct position. 

Sources

All images and directions were provided by our professor, Dr. Cyril Wilson who teaches in the Geography Department at the University of Wisconsin Eau-Claire.  The class is Geography 338: Remote Sensing of the Environment.