Lab 6: Geometric Correction

Goals and Background 

The goal for this lab is to introduce an image preprocessing exercise called geometric correction. This will develop skills using two major types of geometric correction that are typically performed on satellite images as part of the preprocessing activities that come before the extraction of biophysical and sociocultural information from satellite images. Rectification will be used in this lab. Rectification is the process of converting a data file coordinate or grid system known as a reference system.

1. Image-to-Map Rectification: this type of geometric correction utilized a map coordinate system to rectify/transform the image pixel coordinates.
2. Image-to-Image Rectification: this type of geometric correction uses a previously corrected image of the same location to rectify/transform the image data pixel coordinates. 

Methods

Image-to-Map Rectification was the first type of geometic correction used. The USGS 7.5 minute digital raster graphic (DRG) was used to correct the Landsat TM image. Control Points will be used in ERGAS Imagine under the Multispectral tab to perform the correction. The Geometic Model is set to Polynomial and the first order polynomial equation is used. The USGS DRG is also set as the reference map. When you are using a first order polynomial, there needs to be a minimal of 3 ground control points (GCP)

To correct the image, GCP will be placed on both of the images in the same locations. Once four GCPs were placed they need to be adjusted. By looking at the Root Mean Square (RMS) error, it can determine the accuracy between the exact locations on the two images. The industry standard for remote sensing is 0.5 RMS error or below. Since this is an introductory lab, the RMS error needed to be less than 2. The points could be adjusted until the values were below the recommended values. The RMS error total for this exercise ended up being 0.1413 (Figure 1).

Figure 1. Image-to-Map Rectification: Multipoint Geometic Correction with 4 GCP and a RMS error value of 0.1413. 

For part one, even with an RMS error value of 0.1413, there was minimal error in the image that was corrected to begin with. Therefore, displaying the correction was difficult, shown in Figure 2. The final images shown below were resampled using the defaults of the display function to create the new geometrically corrected image. 

Figure 2. Corrected image next to original image with error. 

The second method was Image-to-Image Rectification. The settings for Control Points were all the same except that the equation was changed to a third order polynomial. A third order needs a minimum of 10 GCP. This lab included 12 GCP to get an RMS error total of 0.2027 (Figure 3).

Figure 3. Image-to-Image Rectification: Multipoint Geometic Correction with 12 GCP and a RMS error value of 0.2027.

When the correction was completed, the image was resampled this time using the bilinear interpolation. Using the slide function in ERDAS Imagine, the corrected image can be seen on top of the original image (Figure 4).

Figure 4. Viewer swipe with corrected image and reference image.

Results 

 This lab created a basic understanding of geometric correction. In order to have an accurate analysis, geometrically corrected images are essential. The error may not always be obvious, but when it is zoomed in a difference can be detected. Locating areas that are good for GCP takes practice. But you can never have enough GCP.

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.