Video Resolution Enhancement via Frame Interpolation

In this thesis we propose a novel use of the temporal similarity in frames of a video sequence. Traditionally, video compression algorithms exploit this with interframe coding. Instead, we use the information to enhance the resolution of the video sequence. We use a standard motion estimation technique followed by an analytical method of sub-pixel motion re-estimation that we develop, on each of the blocks in a set of temporally adjacent frames, to obtain a larger data set representing each image in the video sequence. The extra information is then incorporated into the reference image in a useful form by constructing a bivariate spline surface that approximates the enlarged data set, and evaluating this at intermediate points to provide a higher resolution frame. An application GROW, that takes a video sequence as input and produces a higher resolution sequence as output, was developed as proof of concept.