Restoration and identification of image sequences based on detection and estimation of image motion

The availability of digital video has radically changed our expectations of technology, resulting in a large increase in bandwidth use on the internet and an expectation by consumers of high quality images even under adverse conditions such as blur and noise. The use of redundant information in image sequences can either reduce the bandwidth or provide improved restoration over still image restoration. The ability to take advantage of the redundant information, however, depends on the ability to perform good motion estimation. I develop a method to provide improved restoration of an image sequence undergoing global affine motion in the image plane in the presence of additive white Gaussian noise and blur. This type of motion and image degradation is typical of satellite or other aerial surveillance. To achieve this goal, I restore the images using a new edge adaptive method that greatly reduces the presence of edge effects. I also develop a relinearized iterative least squares method to estimate the motion. The results indicate that multi-frame processing can provide a 2--3 dB improvement, while edge-adaptive processing can provide another 3--5 dB of improvement.