Regularización de situaciones de ambigüedad en el registro de imágenes mediante modelos duales

Abstract

Usually, the most critical step involved in image registration is the image matching. A common methodology when estimating the mapping that geometrically relates two images typically consists of two separate and sequential stages: initial feature matching estimation, and regularization for propagating this matching over all image areas. Parametric models representing fuzzy matching regions are proposed in order to support the initial matching. The classical approach has a main drawback, namely the detection of the common features is ambiguous when there is more than one likely matching. To alleviate this problem, the use of dual models to represent the high similarity regions is also proposed in this paper. An averaged POCS (projection onto convex sets) procedure, combined with regularization based on deformable kernels, is used to solve the multiple-choice dilemma. Implementation of these parametrization and regularization steps is described throughout the paper. The proposed approach is tested on a stereo-pair that presents multiple choices of similar likelihood, with successful results.