Lab 6 : Geometric Correction

Lab 6 : Geometric Correction 

Goal and Background 

The main focus of this lab was to introduce the image pre-processing method known as geometric correction.  There are two types of geometric processing used in this lab. The first method used was the interpolation method known as image-to-map rectification.  This process involves converting data file coordinates to another grid and coordinate system.  The map is the reference system and the coordinates are rectified using the map and its coordinate counterparts.  The second method was image-to-image registration.  This method was the same as image-to-map rectification except for the reference system, which is an already corrected image. Both of these methods involved the placement of ground control points and correcting them to minimize the total root-mean-square (RMS) error.  Transformations were applied once the GCPs were placed with a desirable RMS.

Methods 

Part 1 : Image-to-map Rectification  

The first geometric correction method used in this lab was image-to-map rectification. Spatial interpolation was used.  Spatial interpolation is the use of GCP pairs to create a geometric coordinate transformation that is then applied to rectify the location of pixels in the output image with a value from a pixel in the unrectified input image  This was a first order polynomial transformation. A first order polynomial linear transformation fits a plane to the data. This transformation needs a minimum of three ground control points to find a solution.  I activated the multi-spectral processing tools and selected control points. I selected "Image Layer (New Viewer)" in the GCP Tool Reference Setup.  The multipoint geometric correction window opened and I added the reference system (map of Chicago).  I then started adding GCPs.  Again, this is a first order polynomial transformation so we only need three GCPs for the software to find a solution.  I placed four pairs of GCPs just to be safe and it is always good to collect more than the minimum number of GCPs in geometric correction techniques.  After I placed the third pair of GCPs it said "Model Solution is Current".  I no longer need to place two GCPs.  Because the solution is met, it will automatically place a second GCP when I place on on one image, opposed to having to place one on each image like GCPs 1-3. After observing the RMS error, I modified the placement of the points until the total RMS error read 2.0 or lower (Figure 1). The resampling method used here was nearest neighbor.   Finally, I performed the geometric correction by clicking the "compute transformation matrix" and specifying an output location.  

Part 2 : Image-to-Image Registration

This method of geometric correction is very similar to image-to-map rectification, except for the reference system is an already corrected map and it is a third order polynomial transformation, therefore we need a minimum of 10 GCPs for a solution to be met.  I used the swipe tool for both images to evaluate the level of distortion. The next steps for placing GCPs were very similar to part 1, except we placed a total of GCPs and it was a third order polynomial transformation. After applying the transformation, I noticed that the image is much more geometrically accurate than before.  I also made sure the RMS was below 1.0 before applying the transformation (Figure 2). The resampling method used was bilinear interpolation because we are using image-to-image registration.  Finally, I used the swipe tool to evaluate the transformation and it was significantly more spatially accurate than before.   

Results

     

Figure 1

Figure 1 shows the image-map rectification geometric correction method.  You can see the four GCP points as well as the total RMS in the lower right of the window and to the right "(Total)".

   

Figure 2

Figure 2 shows the image-to-image registration geometric correction method.  Notice the 12 GCPs as well as the total RMS in the lower right of the window and to the right of "(Total)".  

Sources

Satellite images are from Earth Resources Observation and Science Center and United States Geological Survey 

Digital raster graphic (DRG) is from Illinois Geospatial Data Clearing House.