Low complexity mode depended local adaptive interpolation filter for next generation video coding

H.264/AVC utilizes the motion prediction with motion vectors at fractional (1/4-pixel). Interpolation filter is utilized to generate the information in the positions between integer positions. Adaptive interpolation filters that adaptively calculate the coefficients by frame have been adopted in KTA tools which were set up as software platform of new proposals for the next video standard. Among several AIFs, EAIF achieved best rate reduction with averagely 7% over all sequences, with complexity doubled. The high complexity restricts its practical applications under current digital computing ability. To reduce the total encoding complexity, mode depended AIF adoption and local macroblock utilization methods are proposed in this paper. It is shown that efficiency of AIF can be related to mode partition distributions which are generated in the first coding pass. By applying different threshold with different QP, the usage of AIF in each frame can be decided. A further complexity reduction is achieved by locally determining the filter usage on each macroblock based on the macroblock characteristics in the second coding pass. Combing two methods, encoding complexity reduction of 40% could be achieved averagely over CIF sequences by means of motion estimation time with negligible penalty to PSNR.