Description of Registration Method for Kapek and
Krekule
Our registration procedure is two stepped in general:
1/ The first step (pre-registration) consists of a feature-based approach
that requires semi-automatic extraction of surfaces of registered objects.
After extraction of the surfaces minimization of their mutual distances is
performed by using chamfer matching algorithm. A function - r.m.s - which
describes the distance of the surfaces is optimized over a parametrical
space of rigid transformations by adaptive simulated annealing that has been
implemented in a stochastic manner.
2/ The second step (fine-registration) is similarity-based and uses mutual
information the value of which is stochastically estimated over a sub-volume
consisting mainly of an object (in order to avoid information-free
background).
Again, the value of mutual information is optimized over a parametrical
space by stochastic adaptive simulated annealing.
Our main effort was paid to acceleration of the whole registration process,
which lead to computation time up to about 30 minutes by using contemporary
PC. In case of quality data sets to be registered - like the data sets
provided by this project - the first step can be omitted.
Return to previous page