Lab 6: Geometric Correction

Goal

The goal of this lab was to introduce us to the preprocessing skill of geometric correction. The lab was structured to develop our skills on the two major types of geometric correction: Image-to-map and image-to-image rectification through polynomial transformation.

Objectives:

1.      Use a 7.5 minute digital raster graphic (DRG) image of the Chicago Metropolitan Statistical Area to correct a Landsat TM image of the same area using ground control points (GCPs) from DRG to rectify the TM image.

2.      Use a corrected Landsat TM image for eastern Sierra Leone to rectify a geometrically distorted image of the same area.

Methods

Image-to-Map Rectification

I opened the provided Chicago.drg.img, which is a USGS 7.5 minute digital raster graphic covering part of Chicago (see figure 1).

Figure 1 This is a USGS 7.5 minute digital raster graphic (DRG) covering part of the Chicago region and adjacent areas. The subset is to show detail.

To rectify the image Chicago_2000 to the Chicago DRG, I used the Multipoint Geometric

Correction tool under Multispectral/Ground Control Points in the ERDAS Imagine interface. The Multipoint Geometric Correction window contains 2 panes. On the left is the input image (Chicago_2000.img), while the reference image is on the right pane (Chicago_drg.img). Each of these panes contains three windows. The top left and top right panes show the entire input and reference images respectively. The other two central top panes shows the areas that are zoomed into on the input image and also that zoomed into on the reference image.

See figure two for a close up view.

Figure 2 The Multipoint Geometric Correction window with the input image (Chicago_2000.img) and reference image (Chicago_drg.img). 

I chose four sets of ground control points (GCPs) to align the aerial image “Chicago_2000.img” with the reference image (“chicage_drg.img”). Though only three GCPs are necessary for a first order polynomial, it is wise to collect more than the minimum required GCPs in geometric correction in order for the output image to have a guaranteed good fit. Figure 3 displays the Multipoint Geometric Correction window. The table at the bottom indicates the RMS (Root Mean Square) Error for the individual points and for the image in total. Notice the RMS error is below 0.5, which means the ground control points are accepted as accurate by industry standards.

Figure 3 The Multipoint Geometric Correction window after placement of the ground control points. The RMS Error is less than 0.5. 

For further explanation of the process, Chicago_drg.img served as the reference map to which we rectified/georeferenced the Chicago_2000 image. Using points from the reference image, a list of GCPs were created that were used to register the aerial image to the reference image in a first order transformation. This works to anchor the aerial image down to a known location; the reference image already had a known source and geometric model. We then used a computed transformation matrix to resample the unrectified data.

The interpolation used was the nearest neighbor method wherein each new pixel in the output image is assigned to the pixel nearest it in the input image.

Matrices consist of coefficients that are used in polynomial equations to convert the coordinates of the input image. A 1st-order transformation was used because the aerial image was already projected onto a plane but not rectified to the desired map projection.

Image-to-Image Rectification

Part two involved doing the rectification process again, this time with two images instead of an image and a map. A third order polynomial transformation was required because of the extent of the distortion, and third order transformations require at least 10 GCPs.

For further explanation of higher order transformations, click here. 

Figure 4 The Image-to-image transformation with twelve GCPS. Notice that the total RMS Error is below 0.05.

Results

Through this laboratory exercise, I developed skills in Image-To-Map and Image-to-Image rectification methods of geometric correction. This type of preprocessing is one that is commonly performed on satellite images before data or information is extracted from the satellite image. The results of the rectification processes can be seen below in figures five and six.

   Figure 5 The image-to-map rectified image of Chicago and the surrounding area.

Figure 6 My rectified image compared to the Sierra Leone image that was used as the reference in the transformation process. The color of my image is washed out, but the orientation appears accurate.

Data sources

Data used in this lab was provided by Dr. Cyril Wilson and collected from the following 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.