| This dissertation addresses the problem of multi-view (multiple camera) image registration in different camera scene configurations and presents two applications of image registration: mosaic image construction and super-resolution (SR) image reconstruction from multi-view images. In this context, we propose a new adaptive filtering framework for local image registration that compensates for the effect of local distortions/displacements without explicitly estimating a distortion/displacement field. The proposed approach aims to register multi-view images in scenarios where the 3-D scene cannot be considered planar, and the accuracy of global parametric image registration techniques proves insufficient. To address the spatially locally varying distortion/displacement field between the multi-view images, we formulate the image registration problem as a 2-D system identification problem with spatially varying system parameters and propose to use a 2-D adaptive filtering framework to identify the locally varying system parameters. In order to register multi-view images with larger perspective distortion, such as larger rotation and change of scale, we propose a new dense multi-view registration technique for wide-baseline video/images that integrates a parametric intensity-based registration method with a sparse set of feature correspondences, based on a locally planar approximation of a nonplanar scene. As opposed to the first approach, the second proposed registration technique provides explicitly the distortion/displacement field.; As image registration applications, we investigate mosaic view and SR view construction. Image registration accuracy directly affects the quality of both mosaic and SR reconstructed images. We present a detailed performance analysis of an SR view construction technique based on projection-onto-convex-sets theory, by taking into consideration a number of factors, including the image registration accuracy. In addition, SR view construction from multiple videos is investigated in the scenario of a dynamic scene. |
