Image Registration

Besides stating the problem of image registration this chapter is built on the following parts : 1) Similarity estimation by level sets. We use the Fast Level Sets Transform (see the chapter on TV minimization) to extract reliable features from both images. To each feature we associate similarity invariants, then we consider pairs of features (F1, F2), where F1 is in image 1 and F2 in image 2, that can match modulo a similarity. We call these pairs correspondences. These correspondences vote for the four parameters of the global similarity. The set of four parameters that gets the maximum number of votes is the estimated similarity. 2) Projective registration. Thanks to a new model for projective deformation, the registration group, we are able, after elimination of the pure projective deformation (i.e. 2 parameters), to reduce the projective matching to a similarity matching.