 |
    |
Using Region Trajectories to Construct an Accurate and Efficient Polyaffine Transform Model
Gang Song, Yang Liu, Baohua Wu, Brian Avants, and James C. Gee
Penn Image Computing and Science Lab, University of Pennsylvania, Philadelphia, PA, USAsonggang@seas.upenn.edu
liuyang@sas.upenn.edu
baohua@seas.upenn.edu
avants@grasp.upenn.edu
gee@mail.med.upenn.edu
http://picsl.upenn.edu
Abstract. In this paper we propose a novel way to construct a diffeomorphic polyaffine model. Each affine transform is defined on a local region and the resulting diffeomorphism encapsulates all the local transforms by a smooth and invertible displacement field. Compared with traditional weighting schemes used in combining local transforms, our new scheme guarantees that the resulting transform precisely preserves the value of each local affine transform. By introducing the trajectory of local regions instead of using regions themselves, the new approach encodes precisely each local affine transform using a diffeomorphism with one or more stationary velocity fields. Experiments show that our new polyaffine model is both accurate and efficient.
Keywords: Polyaffine, Transform, Diffeomorphism
LNCS 7917, p. 668 ff.
lncs@springer.com
© Springer-Verlag Berlin Heidelberg 2013