Volumetric image analysis: optical flow, registration and segmentation

This thesis presents more accurate and efficient methods for volumetric image analysis in terms of Optical Flow, Registration and Segmentation. Firstly, a relationship between the estimation accuracy and the required amount of smoothness for motion estimation from a robust statistics perspective is developed. Next, a fast and accurate non-rigid registration method for intra-modality volumetric images that exploits the information provided by an order statistics-based segmentation method, to find the important regions for registration is presented. Finally, two new methods that improve the accuracy and efficiency of the identification of underlying structures in data that is contaminated with noise and outliers are proposed.