Improved adaptive interpolation filter for H.264/AVC

Motion compensation with adaptive interpolation filters (AIF) was developed to compensate for temporary varieties in the aliasing of video signals and improve the coding efficiency. The AI F achieves better RD performance than common static interpolation filtering by exploiting the statistics in the reference frames' local auto-correlation and the local cross-correlation between the current encoding frame and the reference frames. This paper presents an interpolation filter buffering structure that derives the current encoding frame's interpolation filters from the filters of previous frames. The number of encoding bits for the filter coefficients is reduced by encoding only the differences between the current filter coefficients and the corresponding buffered filter coefficients. Experimental results show that in comparison with the AI F in the current “key technical area” (KTA) reference software, this interpolation filter buffering structure further improves all the test sequences (with up to 2.87%bit rate saving) with negligible computational increase.