Turbo code using adaptive puncturing for transform domain Wyner-Ziv video coding

Abstract

In most DVC architectures presented in the literature, the puncturing mechanism operates according to a periodic pattern that spreads uniformly the turbo generated parity bits and discards all the systematic bits. These latter are replaced by the receiver side information (SI) during the turbo decoding process. The side information is generated by a block based motion interpolation algorithm that can, furthermore, predict whether or not each block is well interpolated. Thus, instead of uniformly dispersing the parity bits over the whole turbo code trellis, they can be directed to the most corrupted regions. If the side information displays regions that appear to be even more highly corrupted, systematic bits can likewise be guided adaptively towards these regions. In this paper, an adaptive puncturing scheme is proposed. A classification of the corrupted parts of each DCT band of the interpolated frame is performed and then an adaptive puncturing mechanism is employed for a more efficient parity and systematic bits repartition. Simulation results show that the adaptive puncturing scheme can provide as much as 20 % of Wyner-Ziv bitrate reduction: this is reflected by a PSNR vs bit rate performance improvement of up to 1.8 dB.