A New Efficient SVM-based Image Registration Method
Daiqiang Peng, DingXue Wu, Jinwen Tian · 2006
A frequently felt difficulty with image registration is the lack of guiding rules to choose a model for unknown geometric distortion. Previous work has concentrated on the use of certain model of mapping function to deal with arbitrarily structured data. The performance of such technique may deteriorate if the model is not well. We consider a general case where a set of models is trained in advance, instead of using one model to register images directly. This technique can find an optimal model for particular deformation. Moreover, central to our approach is that it constitutes a practical implementation of the structural risk minimization principle (SRM) that aims at minimizing a bound on the generalization error of a model, rather than minimizing the mean square error over control points.