Image registration is a key component in image analysis and in problems such as motion estimation or image fusion. In this thesis, discrete optimization is introduced for the task of image registration. A general framework based on random field is derived which allows to represent both linear and non-linear registration as labeling problems where random variables play the role of transformation parameters. Based on this framework, explicit, problem specific models are defined. In order to bridge the gap between discrete labeling and continuous transformations, a novel optimization procedure is proposed based on iterative labeling with successive label space refinement strategies. The procedure is computationally efficient, avoids local minima through large neighborhood search, and yields high-accurate registration results.
«
Image registration is a key component in image analysis and in problems such as motion estimation or image fusion. In this thesis, discrete optimization is introduced for the task of image registration. A general framework based on random field is derived which allows to represent both linear and non-linear registration as labeling problems where random variables play the role of transformation parameters. Based on this framework, explicit, problem specific models are defined. In order to bridge...
»