The purpose of this lab was to introduce a process known as geometric correction. Geometric correction is the manipulation of an image so that it matches a surface projection using GCPs (Ground Control Points) to link a satellite image to a reference image. This process is necessary when a remotely sensed image does not match its planimetric position in the real world. During this lab, two forms of geometric correction were performed. First, an image-to-map first order polynomial geometric correction was performed. This was the simplest form of correction, slightly altering an image using a provided map of the image area. Secondly, an image-to-image third order polynomial geometric correction was performed. This process was far more complex a required the use of a reference image to vastly correct an image far outside of its proper planimetric position.
Methods
Part 1 (Image-to-Map Correction)
First, the image the needed to be corrected, Chicago_2000.img, and the reference map image, Chicago_drg.img, were brought into their own ERDAS imagine viewer. With the viewer containing Chicago_2000.img active, the Multispectral raster processing tool "Control Points" was activated. This started the geometric correction process. The geometric model was selected to be polynomial, and the default GCP Tool Reference Setup setting s were selected. The map image Chicago_drg.img was selected as the reference image, and the default Polynomial Model properties were selected as well, as a first order polynomial transformation was being used. The existing GCPs (Ground Control Points) were removed, as they would not be needed for a first order polynomial transformation, which only requires 3 pairs of GCPs, one of the pair on each image. Three pairs of GCPs were added to the Chicago_2000 image and the reference map image in their corresponding locations, with a fourth pair being added to increase the accuracy of the process. The GCPs were then slightly adjusted to decrease the value of the Total RMS error. This is a measurement of how "off" the program determines the GCP pairs to be. The GCPs were adjusted till a total RMS value of below 2.0 was achieved (Figure 1). The Multipoint Geometric Correction process was then run, creating a geometrically corrected image Chicago_2000gcr.img.
Part 2 (Image-to-Image Correction)
First, the geometrically incorrect sierra_leone_east1991.img was added to a ERDAS viewer, with a reference image sl_reference_image.img being added as well. The geometric distortion in sierra_leone_east1991.img was so great it was determined that a third order polynomial transformation was required to correct the image. This required a minimum of 10 pairs of GCPs rather than 3, as it was fitting a more complex mathematical model to the data. A first order polynomial transformation fits a linear model (y = ax + b) to the data, while a third order polynomial transformation fits a third order cubic model (y = d + cx + bx^2 + ax^3) to the data. The Multispectral raster tool "Control Points" was launched again, with sl_reference_image.img being used as the reference image for sierra_leone_east1991.img. The only change in the setup process from part one, besides the use of different files, was to change the polynomial order to 3 in the Polynomial Model Properties. Similarly to part one, the existing GCPs were deleted and replace with 12 pairs of corresponding GCPs on the reference and sierra_leone_east1991 image. The points were then slight altered, until the total RMS value was below 1 (Figure 2).
Results
Part 1
The resulting geometrically corrected image Chicago_gcr.img, at an initial glance, shows little change from the original satellite image Chicgo_2000.img. This is because the first order polynomial transformation used on this image only slightly shifted the original image so that the surface features were correct in regards to their position on the reference map (Figure 3). The amount of geometric distortion seen in the original image is relatively minor, and the image only needed to be altered slightly to match the real world positioning of surface features.
In comparison to the original sierra_leone_east1991.img, the geometrically corrected sl_east_gcc.img has been mostly corrected in regards to its its planimetric position in the real world. When overlayed on the reference image and analysed using the swipe tool, the corrected image largely matches the reference image (Figure 4). In comparison, the original image, when overlayed on the reference image, is shown to be position far above where is should be. While sl_east_gcc.img is now largely geometrically correct, it is crucial to not that the image is still not perfect. In particular, the top-left corner of sl_east_gcc.img is stretched slightly beyond the boundaries of the reference image, while the bottom left corner does not quite reach the edges of the reference image. In the future, these errors could be largely corrected by increasing the number of GCPs between the original and reference images and/or by decreasing the individual and total RMS values of each GCP pair.
Sources
Earth Resources Observation and Science Center. In United States Geological Survey. Retrieved November 17, 2016, from http://eros.usgs.gov/usa
In Illinois Geospatial Data ClearingHouse. Retrieved November 17, 2016, from https://clearinghouse.isgs.illinois.edu
Wilson, C. (2016). Lab 6: Geometric correction. Eau Claire, Wisconsin.
No comments:
Post a Comment