A new algorithm for the computation of the group logarithm of diffeomorphisms
Abstract
There is an increasing interest on computing statistics of spa- tial transformations, in particular diffeomorphisms. In the Log-Euclidean framework proposed recently the group exponential and logarithm are essential operators to map elements from the tangent space to the man- ifold and vice versa. Currently, one of the main bottlenecks in the Log- Euclidean framework applied on diffeomorphisms is the large computa- tion times required to estimate the logarithm. Up to now, the fastest ap- proach to estimate the logarithm of diffeomorphisms is the Inverse Scal- ing and Squaring (ISS) method. This paper presents a new method for the estimation of the group logarithm of diffeomorphisms, based on a se- ries in terms of the group exponential and the Baker-Campbell-Hausdorff formula. The proposed method was tested on 3D MRI brain images as well as on random diffeomorphisms. A performance comparison showed a significant improvement in accuracy-speed trade-off vs. the ISS method.
Domains
Other [cs.OH]
Origin : Files produced by the author(s)
Loading...