High Quality Spatial Interpolation of Video Frames using an Adaptive Warping Method

The scenario we address in this paper is high quality interpolation of 'talking head' lecture video. It is often desired to enlarge streaming media of this type with frame size NtimesM to a higher spatial dimension of, let's say, 2Ntimes2M. Simple interpolation methods are sometimes preferred because of complexity constraints in processing. But even in the absence of such constraints, the quality of the image degrades as the image size is increased, typically resulting in a blurred version of the original. Here we report on a hybrid interpolation approach for high quality frame size expansion for lecture type video. The method assumes that we retain one high resolution reference frame of size 2Ntimes2M, which we use as a base to derive the other frames. Two variants are considered for deriving the high frequency frame content. Both achieve significant improvement compared to conventional interpolation, resulting in frames that are sharper and clearer