Lab 6 - Geometric Correction

Goals and Objectives
The goals of this lab are to learn the different methods of image preprocessing or geometric correction. There are two main methods of geometric correction, image to map rectification and image to image registration. Part one highlights image to map rectification, using a topographic map of Chicago and the satellite image. Part two involves image to image registration using two satellite images of Sierra Leone. The goal of this exercise is to use ground control points on both sets of images to rectify the planimetric position of the study satellite image. An indiciator of high quality geometric correction is the root mean square error (RMSE). A RMSE below .05, or within half a pixel, will result in a higher quality geometric correction.

Part 1
In this part, a topographic map was created of the Chicago area and digitized to create a digital raster graphic (.drg). Digital rasters can serve as a planimetric base with which to acquire ground control points for image to map rectification. In this part's example, we are operating at a first order polynomial model. 1st order polynomial involves acquiring at least 3 ground control points in order to rectify the image. The method used to rectify the image was nearest neighbor which uses the closest pixel to estimate the brightness values of the rectified image.

Figure 1 Using Multipoint geometric correction to manually enter ground control points from a reference topographic digital raster graphic. RMSE total of .0003.

 Part 2
This part of the exercise used image to image registration to geometrically correct the image. A satellite image that was previously corrected was used to process the newly acquired satellite image. In this example, unlike part 1, the third order polynomial order was used, in which a minimum of 10 points are required to correct the image.  Bilinear interpolation was also used to produce the corrected ouput image, which uses the 4 nearest pixels to estimate the output image's brightness values.

Figure 2 Bilinear interpolation multipoint geometric correction done in ERDAS Imagine of Sierra Leone 1991. RMSE total of 0.0127 with 13 ground control points.

Conclusion
It was interesting to learn in this lab and corresponding lecture about how satellite images need to be corrected due to systemtatic and non systematic errors, such as equipment operation or natural earth tendencies. Learning to use ground control points was fascinating, I prefer the image to map method but I found it interesting to note that the higher the polynomial model and the more ground control points, the harder it is to reduce the root mean square error.