Temporal video interpolation with nonlinear motion estimation

In video sequences, some techniques such as frame repetition or linear interpolation for reconstruction of skipped, lost or nonexistent frames within video streams, in an attempt to generate more fluid video sequences and slow motion playback, result in undesirable artifacts, noise and unnatural effects. An effective technique called motion compensated temporal interpolation (MCTI) is used to interpolate well-reconstructed video frames in the time domain. However these techniques focus on the assumption that the objects in the video exhibit linear motion characteristics. In this paper, we propose a new temporal video interpolation that can also model nonlinear spatio-temporal motion of the objects. Results indicate that the nonlinear motions approach described here is superior to the linear-based methods that were previously studied.