Geolocating Images
Most satellite images are now shipped with location information, or they are located by sensor position and lens direction, which usually is precise enough for connection with other images or map data. However, some have no interal location or device pointing info. To “geolocate” an image, we compare it to a map or another image using one of several processes.
- Registration – co-locating two images that may or may not be in proper coordinates. Clicking the same point on both registers one to the other.
- Georeferencing – puts the image data into geographic coordinates associated with a projection and datum. “Ground control points” are used to locate the same feature on map and image.
- Orthorectification – both georeferences image data and corrects for distortion due to terrain elevation (layover). It requires a DEM and information about the camera or sensor. The ground control points are 3D in this case.
The rectification should be done AFTER the classification or other image analysis, because the process usually degrades the quality of the original image by resampling or otherwise merging values in cells.
The process has three basic steps.
- Finding control points
- warping the image using a polynomial transformation or other schema. The polynomial “order” determine how much distortion can be admitted into the warping model
- a “linear” model allows only shifting in x and y, rotation, or skew in scale in x and y
- a “polynomial” or “2nd order” transformation allows the scale to change with distance in x and /or y, plus the above
- a “cubic” transformation allows the above plus localized distortion due to the lens (not common in satellite image)The Root Mean Square Error of the warping model is
where N is the number of points and y is the location you picked and x is the location predicted for that point using the other points you picked.
The RMSE expresses the degree of error in all of the points. In general is it best to use the LOWEST order transformation that you can get away with that produces an RMSE approaching 1 pixel.
– ArcGIS Pro has other options that you should investigate.
- resampling the image to create new pixel in the geolocated reference space (See Bolstad’s Figure 4.30)
- nearest neighbor selects the closest cell to the correct geographic location (computationally easy, but can result in shifts in linear features and edges)
- bilinear interpolates the value for the DN in the 4 cells closest to the output cell in x and y (extreme data are lost, but image is “smoothed”)
- cubic convolution uses an array of 4 cells in both x and y fit to a polynomial (tougher for your CPU)
see this page or you text for images on these different methods
https://www.nrcan.gc.ca/maps-tools-publications/satellite-imagery-air-photos/remote-sensing-tutorials/image-interpretation-analysis/pre-processing/9403
http://seadas.gsfc.nasa.gov/help/general/ResamplingMethods.html
Note: if you are not going to classify or interpret the data first, and will be resampling original images, you should only use nearest neighbor, because the others will average out the cell values and mute the original image.
much of this from www.coastal.uno.edu/coastal/high-edu/cs4096/lectures or the Univ AZ Remote sensing tutorials