Image Registration

Image registration takes an image like an aerial photograph or scanned map, and assigns coordinates to some points so that the computer can calculate the coordinates for every point.  If the process is complicated, the program might Reproject the data  so that it has a simple geometry.

In the simplest case, the coordinates for three points will define a translation and rotation.  This is an affine transformation.

The following conditions will limit the effectiveness of such registration, and require more complicated algorithms:

• Lens distortion (poor optics), either in a camera or scanner.
• Off-nadir viewing geometry.  This distorts the view of the camera from what should be a rectangle, into a parallelogram.
• Terrain displacement in steep areas for imagery.
• For scanned historical maps, low quality of the original map.
• Paper distortion for old paper maps.  Mapping agencies use mylar sheets for archival purposes, and many maps and charts use high quality paper which does not show the differential stretching of cheaper paper.

The term "rubbersheeting" refers to a stretching or warping of the image to bring it into the correct geometric location.

Triangulation registration maps points on one map to another, computes an optimal set of triangles, and then distorts the map to obtain a correct geometry. 

• The most robust solutions use GDAL, which you must have installed.
  • You must know the projection and datum used in the map, or make an informed guess.  If the scale is such that the datum shift will not be noticeable on the map, you can ignore that factor.
  • Use QGIS.  Follow the tutorial at http://www.qgistutorials.com/en/docs/georeferencing_basics.html
  • Use MICRODEM.

Database with image registration points Image coordinates came from the location where the mouse was clicked.  The origin (0,0) is the upper left corner of the image, the computer graphics convention.  The user will not interact with these, other than not panic about their reference system. Lat/long coordinates were entered from the keyboard. UTM coordinates for the current zone, converted from the lat/long values entered.

• The steps in both cases, and in others such as ArcGis, involve:
  • Selecting points on the image, and giving the program their coordinates.  The can be via the keyboard, or by clicking on an already registered map in the GIS.  All point selection must be done at the full resolution of the maps, for the maximum accuracy of the coordinates.
  • If the map has a grid or graticule
    • The four corner points offer a good starting point.  For a large scale map, those four points will often provide an adequate registration.
    • If the map has either a full grid/graticule, or with ticks in the center of the map, those points will allow a more precise registation than allowed by just the corners.  On a common quadrangle map, there will be 16 points in the graticule on the map.
  • In optimal circumstances, two points will suffice, but in general you need three and should select at least four.
  • The registration points must be spread throughoug the map, and cannot be collinear.
  • Having the program register the image, which will create a new raster.
  • Verify the results.

Last revision 12/4/2017