Adaptive Interpolation Filter for Motion Compensated Hybrid Video Coding

Common hybrid video coding standards such as H.263, MPEG-1,2,4 are based on motion compensated prediction. For the motion compensated prediction purpose, displacement vectors with fractional-pel resolutions (e.g. 1/2-pel or 1/4-pel) are used. In order to estimate and compensate fractional-pel displacements, the image signal has to be interpolated. The most popular and very simple interpolation method is the bilinear interpolation. Recent approaches like the MPEG-4 ACE-profile and H.26L use Wiener interpolation filters. Up to now these filters are based on filtercoefficients that are invariant. The same filter is used for all sequences and for all images of a sequence. In this paper an adaptive interpolation filter is introduced in order to improve the coding efficiency. The adaptive interpolation filter is based on filter-coefficients that are estimated for each image. Thus, the influences of the displacement estimation error and the aliasing components, which are deteriorating the motion compensated prediction are reduced. Analytical calculations and experimental results show a dependence between the variance of the displacement estimation error and the filter characteristic. Compared to the H.26L codec the adaptive interpolation filter improves the PSNR up to 1.0 dB for a sequence. For single images with high displacement estimation errors, gains up to 1.5 dB are obtained.