The capturing of multiple low-resolution images taken of the same scene results in a distortion between each image. Image registration is the process of determining this distortion. This information is then used in the super-resolution process to create a set of simulated low-resolution images. The differences between these simulated images and the observed images are then used to iteratively update an initial estimate of the high-resolution image. Successful super- resolution is dependent on accurate image registration. In this thesis, we examine the hypothesis that the visual quality of a reconstructed high-resolution image improves when accurate image registration is achieved.
In the first part of this thesis, we examine the image registration process in detail. Both picture and text images are registered using two algorithms. The first registration algorithm based on an optimization approach whilst the other is based on the RANSAC algorithm. We find that the optimization approach is severely hampered by higher degree transforms such as affine transforms. This is attributed to the increased number of parameters requiring optimizing.
In the second part of this thesis, we focus on the super-resolution process. Numerous experiments were conducted to test our original hypothesis. The first experiment involved reconstructing an image when perfect registration was achieved, and comparing the results to when the RANSAC algorithm was employed. The results suggested that the visual quality of the reconstructed images were higher for perfect registration. We also found that the visual quality of reconstructed images was higher when images were registered using the RANSAC algorithm, as compared to an optimization approach.
Keywords: Image registration, super-resolution, image enhancement, image alignment, image reconstruction.