Adaptive Interpolation Filter Using Correlation for Inter Prediction

In video coding, interpolation filters play an important role in motion-compensated prediction. In Versatile Video Coding (VVC), adaptive motion vector resolution (AMVR) that makes use of fractional samples by 8-tap Discrete Cosine Transform-based interpolation filter (DCT-IF), 7-tap DCT-IF, or 6-tap Smoothing interpolation filter (SIF) was adopted. This paper proposes an adaptive interpolation filter to improve motion-compensated prediction. A 12-tap DCT-IF is designed to improve the filter response in the high frequency range and applied by a proposed filter selection method using correlation that considers the frequency characteristics of reference block found from motion estimation. Experimental results show that the proposed method achieved the overall Bjontegaard Delta (BD)-rate gains of -0.27%, -0.12%, and -0.01% for Y, Cb, and Cr components, respectively, under the Random Access (RA) configuration compared with VVC.