Goals and Background
This lab introduces geometric correction. Both image to map rectification and image to image rectification will be performed on two different images. In both cases, spatial interpolation will be used with the use of GCPs from a reference image or map to change the x,y location of the pixel, and intensity interpolation will be performed by facilitating resampling to generate the relocated pixels brightness values. Image to Map rectification is detailed in part 1, and image to image rectification is detailed in part 2.
Methods
Part 1: Image to Map Rectification
For this part, a reference map is used to rectify a distorted image. The reference map in this part is a 7.5 minute raster of the the Chicago area, and the distorted image is from the Landsat TM satellite of the Chicago area.
Image rectification was done by first inserting the reference map and distorted image in different viewers in Erdas. Then, with the viewer with the distorted image being active, the control points button under the multispectral tab was used to bring up the Multipoint Geometric Correction window. When setting the window up, a first order polynomial was used for the GCPs, and the reference map was brought in, and all of the other defaults were accepted.
The Multipoint Geometric Correction window contains two panes. One for the distorted image and the other for the reference map, This can be seen below in figure 3.0. The distorted image is in the left pane, and the reference map is in the right pane.
 |
| Fig 3.0: Multipoint Geometric Correction Window |
Then, 4 GCPs were created using the Create GCP button. These GCPs were placed so that they were spread out in both images. The locations of the GCPs were made sure to be located in the same location in the distorted image and in the reference map. The placement of the GCPs can be seen below in figure 3.1.
 |
| Fig 3.1: GCP Placement |
Part 2: Image to Image Rectification
This part is similar to part one, but instead of using a map as the reference layer, an image will be used to rectify a distorted image. Also, The images for this part are of eastern Sierra Leone taken by the Landsat TM satellite. The distorted image in this part is more distorted than in part 1.
To rectify the distorted image, first, the distorted image was put into a viewer in Erdas. Then, the Multipoint Geometric Correction window was brought up by clicking on the Control Points button under the mulispectral tab. This time, while setting up the window, the order of polynomials was changed to 3rd order. Doing this increases the number of GCPs needed to make the model current from 3 to 10. Also, the reference image was brought in, and all of the other default settings were accepted.
Then, 12 GCPs were created using the Create GCP button. These GCPs were placed so that they were spread out across the images. The reason why the GCPs are spread across the image was because the distorted image needs to be pinned down to the correct location in various locations of the reference image so there is a good coverage of the whole image. Otherwise, if the GCPs are located all right next to each other, the rectified image will not be spatially accurate. The locations of the GCPs were made sure to be located in the same location in the distorted image and in the reference image. The placement of these GCPs can be seen below in figure 3.2. The distorted image is in the left pane, and the reference image is in the right pane.
 |
| Fig 3.2: GCP Placement in the Multipoint Geometric Correction |
After the GCPs were added, they were altered until the total RMS error was .1446. Then, using the Display Resample Image Dialog button, intensity resampling was performed on the distorted image using the bilinear interpolation technique. The bilinear interpolation technique is used because this image is more distorted than the image was in part one and because bilinear interpolation is more spatially accurate than nearest neighbor interpolation. This image needed the extra help in order to make the output accurate. This resampling generates a rectified image which is more spatially accurate than the original distorted image.
Results
The results of part one is a rectified image which can be seen below in figure 3.3. This figure is a short video which utilizes the swipe tool to show how accurate the rectified image compared to the reference map. This video shows that the rectified image is very accurate as features line up almost perfectly with the map. This can be seen by looking at the Lake Michigan shoreline and also by how rivers appear to be stacked right on top of each other in the images. To increase the video quality one can watch the video in full screen mode.
Fig 3.3: Rectified Image Compared to Reference Image
Figure 3.4 shown below is the rectified image by itself. This is shown here because the image is a little more clear here than what is shown in the video.
 |
| Fig 3.4: Chicago Rectified Chicago Image |
Fig 3.5: Sierra Leone Rectified Image Compared to Reference Image
Figure 3.6 shows a clearer picture of the rectified Sierra Leone image. Overall, the rectification did a decent job of making the distorted image more spatially accurate. Perhaps with more GCPs and a lower RMS error, a more spatially accurate output could be created.
 |
| Fig 3.6: Rectified Sierra Leone Image |
Sources
Earth Resources Observation and Science Center. Satellite Images
Illinois Geospatial Data Clearing House, Digital raster graphic (DRG)
United States Geological Survey. Satellite Images
Wilson, Cyril, 2017. Geometric Correction retrieved from
No comments:
Post a Comment