Low Complexity Mode Depended Adaptive Interpolation Filter For Next Generation Video Coding

Fractional precision motion estimation is implemented in H.264/AVC. To furtherimprove coding efficiency, the on-going video standard HEVC utilizes the adaptiveinterpolation filter (AIF) which adapts the coefficients frame by frame. AIF can reducebitrates significantly; however its complexity increases a lot.In this thesis, we focus on reducing AIF complexity while maintain its highcoding efficiency. Three methods are proposed in this thesis. The first proposal focuseson early elimination of AIF before the second coding pass in the encoding process.This proposal explores the correlation between AIF performance and partition modedistributions. By setting a threshold based on small partition mode occurrence, the AIFis disabled in sequences which no coding efficiency could be achieved after applyingadaptive filtering. The second method applies to the coefficients calculation process,by skip calculating the skip macroblock in every frame, small amount of complexitycan be saved in coefficients finding process. Combining proposal one and two,averagely 59% encoding time can be saved in CIF, up to 25.6% and averagely 16%encoding time can be saved in 1080p sequences. The third method is applied to firstencoding pass, unlike proposal one, this method can be applied in every sequenceswhile providing constant reduction of complexity under all conditions. This methodsimplified the first coding pass base on the observation of the effect of first pass codingto the whole encoding process. The experiments shows that, for 1080p sequences, thismethod can constantly reduce around 20% encoding time and with slightlyimprovements on coding quality. Thus this method introduces no penalty to codingefficiency while reducing the complexity constantly.