Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
533988 | Pattern Recognition Letters | 2013 | 5 Pages |
•Mapping models used in image registration should form a group.•We show that so-called weak affine transform is neither invertible nor closed.•We show there exist affine subgroups which can serve as restricted affine mapping instead of weak affine transform.
In the recent paper (Li and Zhang, 2012) a mapping model of weak affine transform (WAT) was proposed for registering images of flat scenes. We show that such a model may cause inaccuracies when registering images with non-uniform scaling differences, even if they seemingly behave according to this model. Mathematical reason of this is that WAT is neither invertible nor closed. We explain why mapping models forming a group should be preferably used for registration and show some examples of models belonging to affine subgroups.