Restoration of dynamic image sequences: Resampling and high-resolution reconstruction

With the increasing applications of dynamic image sequences for scientific data acquisition, evaluation, analysis as well as entertainment, it is quite necessary to process such data to increase the usefulness of the information provided by the images. For example, one may want to change an image sequence or video from one format to another, remove artifacts due to blurs and noises in the image sequence or get the details from a set of image frames, etc. In this thesis, we address the problems which occur when one desires such improvements from given low-resolution (undersampled) and possible distorted dynamic image sequences.; First, an efficient block-based resampling method is developed for discrete signals. Significant reduction of the complexities is achieved by direct use of transform coefficients in the resampling process.; Secondly, a high resolution image synthesis method is proposed from a sequence of blurred, low resolution, noisy and aliased image frames. In this development, it is assumed that the image sequence is static--no moving objects in the sequence. The result can be extended to dynamic image sequences, which is described in Chapter 5. Recursive and parallel structure of the algorithm make the computation very efficient.; An image registration problem is then studied. The proposed algorithm provides the spatial shift information required in the high resolution reconstruction algorithm. This registration method can estimate the shift between two static aliased image frames to subpixel accuracy.; Lastly, an algorithm for high resolution reconstruction of dynamic image sequences is developed in Chapter 5. The research results presented in Chapter 2 through Chapter 4 are combined under block-based processing. The block-based decomposition of image frames is employed in order to reduce the nonlinearities due to nonuniform motions in a dynamic scene.; The algorithms developed in this dissertation have been tested and verified through extensive computer simulations. Further improvements of dynamic image sequence restoration may be possible by use of more elaborate object-oriented image segmentation technique for image decomposition. The results of this research will have applications such as up-conversion of conventional video signals to HDTV format, high resolution video printing, as well as other scientific multiframe image information analysis applications.