Motion and Aliasing Compensated Prediction

TNT members involved in this project:

Previous motion compensated hybrid video coding standards such as H.263, MPEG-1 and MPEG-2 (H.262) are based on a fractional-pel displacement vector resolution of 1/2-pel and a Bilinear interpolation filter. The newest standard H.264/AVC allows for 1/4-pel resolution with a fixed Wiener filter applied to half-pel positions and a Bilinear filter applied to quarter-pel positions. In general, motion compensation is disturbed by aliasing components contained in an image signal, thus limiting the prediction accuracy obtained by motion compensation.

The motion and aliasing compensated prediction aims at reducing the aliasing, quantisation and motion estimation errors, camera noise, etc. More accurate motion compensation exploiting image contents adaptively, results in less prediction error to be coded, thus making video source coding more efficient.

Aliasing reduction is achieved by applying a Wiener interpolation filter, which is indepently calculated for each frame. In order to enable analytical calculation of the filter coefficients, a non-separable 2D filter scheme is applied. For every fractional-pel position to be interpolated, an individual set of 2D filter coefficients is determined. As a result, a coding gain of up to 1.0 dB for HDTVsequences and up to 0.6 dB for CIF-sequences compared to the standard H.264/AVC is obtained. This approach became the start point for the current exploration software of VCEG. The exact coding gains results for an extensive test set can be found here. Furthermore, in order to show that the approach does not decrease the visual quality, here some results:

Sequence Config file Data rate Y-PSNR Download
City, 720p, 60Hz, reference encoder_city_std.cfg 2178.492 kbit/s 34.549 dB
City, 720p, 60Hz, AIF encoder_city_aif.cfg 2035.042 kbit/s 34.737 dB
Foreman, CIF, 30Hz, reference encoder_foreman_std.cfg 223.489 kbit/s 34.087 dB
Foreman, CIF, 30Hz, AIF encoder_foreman_aif.cfg 202.504 kbit/s 34.217 dB

Show all publications
  • Deyao Zhu, Marco Munderloh, Bodo Rosenhahn, Jörg Stückler
    Learning to Disentangle Latent Physical Factors for Video Prediction
    German Conference on Pattern Recognition (GCPR), September 2019
  • Yuri Vatis, Jörn Ostermann
    Adaptive Interpolation Filter for H.264/AVC
    IEEE Transactions on Circuits and Systems for Video Technology, Vol. 19, No. 2, pp. 179 - 192, February 2009
  • Yuri Vatis, Jörn Ostermann
    Syntax of adaptive filter coefficients in the KTA reference model
    ITU-T Q.6/SC16, doc. VCEG-AF09, San Jose, USA, April 2007