Non-rigid image registration regularization, algorithms and applications

Image registration aligns multiple images of the same scene so they can be compared and analyzed, yet automated correspondence detection—especially under non‑rigid deformation—remains challenging. This dissertation presents several contributions to non‑rigid image registration. First, a new intensity‑based similarity measure, Residual Complexity (RC), is introduced to handle spatially varying intensity distortions common in microscopy and MRI. Second, a novel rigid and non‑rigid point‑set registration method, the Coherent Point Drift (CPD) algorithm, is developed to jointly estimate point correspondences and transformations. Finally, a fully automated motion‑estimation framework for 3D+T echocardiography is proposed, incorporating ultrasound‑specific similarity measures and biomechanical constraints to generate dense deformation fields for quantifying myocardial strain and torsion. Validation using implanted marker data demonstrates the accuracy and clinical potential of the approach.