The popularization of information-sensing devices and rapid development of data storage and computing capability have launched the ongoing data explosion, and the value of such information assets has been widely acknowledged. Data often come in high volume, variety and velocity, and people are striving to mine useful information and new knowledge from the growing data sets for various domain applications. The fusion of heterogeneous types of data becomes increasingly essential for analysis and decision making in data incentive industry, especially in healthcare industry. Image registration, as a branch of data fusion technique, is the process of determining the geometric transformation that relates correspondences in images. As the image registration result is often used as the input of further analysis or process, it is crucial to establish the image correspondence through an accurate and confident registration method. However, due to the diverse characteristics of images, there are inherent uncertainty issues associated with various aspects of the registration process. The source of such uncertainty issues could be divided into two categories: input images and registration algorithms. This thesis provides three registration methods with the consideration of uncertainty issues the of registration processes. Our major contributions include: 1. A topology-guided deformable registration (TDR) framework to deal with the image correspondence uncertainty issue in the derivation of deformation direction. 2. A novel laminar flow (LF) model with an analogy to laminar flow regime from fluid dynamics to simulate the derivation of registration transformation. 3. A multi-view collaborative learning based image registration (MVCIR) framework to tackle the uncertainty of deciding reliable feature space for the correspondence inference.