An Iterative Method for Frame-Level Adaptive Wiener Interpolation Filters in Video Coding

In this paper we present an iterative method for improving frame-level adaptive Wiener filtering in video coding. Our method iteratively computes Wiener filter coefficients using the motion field, macroblock partitioning and best coding modes, which are in turn updated in the motion estimation and compensation process using the computed filters. To control the computational complexity of the iterative method, we also propose a termination condition using the mean Euclidean distance of the filter coefficients between consecutive iterations. With extensive computer simulations, we found that our proposed iterative Wiener filtering method produces significant bit rate saving (3.4% on the average) while maintaining the same level of peak-signal-to-noise-ratio (PSNR), compared to the JM 10.1 reference software for H.264/AVC. Most of the performance gain comes from a few iterations at the very beginning. As the number of iteration increases, the further bit rate reduction is less than 1%. Therefore, to achieve the best balance between computational complexity and performance gain, we believe computing Wiener filter coefficients using just 2-3 iterations will suffice