A bayesian estimation of parameters for point local patch registration

The local color distribution centered on feature point is usually used for image registration, in which the estimation for transformational parameters is the core problem. As the reported methods can not suit for large image deformations and are on the assumption that transformational parameters changes are followed by linear or Gaussian distribution so that the estimation process is easy to get into local extremum. So a Bayesian estimation of point local patch registration parameters is presented in this paper. This method apply MCMC algorithm to solve the posterior probability distribution problem and obtain the optimal parameters. The test results prove this method can perform point local patch registration under large image deformations accurately and successfully coped with target scale or orientation variations in video tracking.