Video coding with cubic spline interpolation and adaptive motion model selection

We propose a block based video coder with multiple motion models. The coder is designed such that each block can be predicted using the motion model with the best rate-distortion performance. In such a scheme, there is a significant impact on the interpolation, motion compensated prediction structure and the design of motion vector predictors, because neighboring blocks could have been predicted by different models. We employ cubic splines to perform interpolation and use the spline framework to estimate the translational shift and warping of blocks. We address the problem of adaptive model selection and enhance the motion vector predictors to account for multiple motion models. Bitrate savings up to 11.7% are observed for the tested sequences compared to an improved H.264/AVC reference.