This software uses the affine transformation to register two images. Consecutive frames in a video captured
by a distant camera of a rather flat scene can be registered by the affine transformation. The registration
software is automatic. The user only provides the names of the two image files to be registered. The program
will then register the images and display the result. The program selects a set of landmarks in each image,
determines the correspondence between the landmarks, calculates the transformation parameters from the
correspondences, and registers the images using the obtained affine transformation.

Assumptions: It is assumed that
  • parallel lines in the scene remain parallel in the images
  • the images are in the same modality,
  • at least 50% overlap exists between the images.

If one or more of these assumptions are violated, the program may register the images approximately, or
report a failure.

Modes of operation: The program operates at three different modes.
  • Default mode: This is a fast initial mode used to register the images quickly. For some images, this is
    sufficient.
  • Redo mode: For images with considerable intensity differences, it may be necessary to redo the
    registration using a larger number of landmarks. The user presses a Redo button after viewing the
    registration result of the default mode.
  • Refinement mode: In this mode, the positions of the landmarks are more accurately determined by
    template matching and the registration is repeated. Examples of image registration by this software are
    given below.
(a)
(b)
(c)
(d)
Fig. 1. (a), (b) Two aerial images taken from slightly different views. (c) Registration using the default mode. This
takes about 5 seconds. (d) Refinement of the registration. Affine transformation is an approximation to the
projective transformation, an approximation that may contain large errors when view-angle difference between
the images is large. Affine registration is suitable for registration of satellite images as demonstrated below.
(a)                                                                                      (b)
(c)                                                                                      (d)
Fig. 2. (a), (b) Two Landsat images with small view-angle differences. These images are courtesy of NASA. (c)
Initial registration. (d) Refined registration by this software.
To obtain a license for this registration software, please follow the link =>
2-D auto affine registration
Image Registration and Fusion Systems