Low Complexity Image Registration Techniques

Abstract

This dissertation explores fast and efficient image registration techniques for different innovative applications in the diverse areas of video coding and medical image analysis. One important recent application of image registration is for the estimation of global motion parameters for object-based video coding. To estimate the global motion parameters, a sophisticated motion model needs to be integrated into the estimation process. The use of these models produces increased computational complexity in the optimization techniques. A part of this thesis presents a new adaptive low-complexity image registration technique to deal with this limitation. The unique features of the proposed algorithm are: replacement of arithmetic operations with logical operations, complete elimination of the Hessian matrix calculation and an adaptive step-size procedure for each gradient descent iteration. Next, a novel 3D/2D registration technique is proposed for 3D/2D registration of computed tomography (CT) to single plane X-ray fluoroscopy. The application of fast and efficient 3D/2D image registration can be enormously helpful for different clinical purposes, such as image-guided surgery and the kinematic analysis of bones in knee and ankle joints. A limitation of this approach is the need to recalculate the voxel values in the 3D volume for every iteration of the registration procedure prior to generating a digitally reconstructed radiograph (DRR). In this thesis, a new multi-phase 2D-3D image registration algorithm is proposed which employs partial 3D volumes to estimate out-of-plane rotations. In the proposed approach only one full 3-D update is used. As a result of reducing the number of 3-D updates, the proposed approach reduces the time required to perform the registration. Finally, a fast and accurate 3D/3D single-mode image registration approach for rigid-body transformations is proposed for CT and magnetic resonance imaging (MRI) images. Important application areas exist for single-mode 3D/3D registration, such as: treatment verification by comparison of pre- and post-intervention images and the monitoring of the abnormal growth of tumors or organs. The unique feature of the proposed algorithm is the estimation of the six rigid-body transform parameters by using a new multi-cube based approach in a hierarchical fashion. The registration precision, success rate and speed up of the proposed algorithm compares favorably with other state-of-the-art techniques.