The library is designed as a set of techniques to allow realization of a variety of correspondence-based registration algorithms.
Components of the framework include initializers, feature sets, correspondence generators, robust transformation estimators, and convergence testers. Multiple feature types may be used either simultaneously or sequentially, and multiresolution is handled naturally.
class rgrl_feature_based_registration
- the engine that represent the framework
class rgrl_feature
- basic primitives of correspondence-based registration
- pre-computed or constructed dynamically
class rgrl_feature_set
- a collection of features
- with representations suitable for spatial queries such as finding the nearest neighbors
class rgrl_initializer
- method to generate the initial transformation
class rgrl_matcher
- generating correspondences from the two image feature sets.
class rgrl_estimator
- estimating the spatial relationship that maps the moving image to the fixed image using the matches
- transformation parameters stored in the rgrl_transformation of the same model
class rgrl_scale_estimator and rgrl_weighter
- for robust transformation estimation in the presence of outliers/mis-matches
class rgrl_convergence_tester
- testing whether the estimation has converged to a stable solution or is oscillating.
