Stochastic search for 3-d shape registration

Global optimization is fundamental in many problems such as 3-d registration, which is critical for Non-Destructive Testing. Global optimization using stochastic search has been popular in human motion capture. These methods use annealing and construct a sequence of particles that approximate a likelihood, whose global mode is searched by the optimization procedure. Treating the minimization objective as the negative log likelihood ensures that the global mode of the likelihood corresponds to the minimum of the cost function. Hardly any research has gone into studying how well these global optimization tools can be adapted for 3-d registration, which require the optimization of non-convex objectives that would result in a multi-modal likelihood. In a previous work, we applied these methods for 3-d bilateral symmetry plane estimation. In this paper, we apply stochastic search techniques to the problem of 3-d shape registration.