Drift analysis and drift reduction for multiresolution hybrid video coding

摘要

The prediction error instead of the original signal has to be scaled for efficient multiresolution hybrid coding. After motion compensation, the prediction error is scaled by low-pass filtering and subsequent subsampling. However, causal prediction becomes a problem when only the coded low frequency information is used to derive low resolution images at a base layer decoder. In general, a drift between the predictors of the multiresolution encoder and the base layer decoder occurs. This drift results in an additional reconstruction error of the low resolution images. It is due to aliasing which is introduced by subsampling and cross talk which is introduced by motion compensation in combination with low pass filtering. Correspondingly, a two step drift reduction method is proposed. In the first step, Wiener filtering reduces drift due to aliasing and in the second step, overlapped block motion compensation reduces drift due to cross talk. Experimental results confirm that the proposed method substantially reduces drift by 7.9 dB on the average compared to a reference method which consists of bilinear interpolation filtering in the first step and conventional block motion compensation in the second step. The drift reduction method can also be used to improve the prediction of the low frequencies at a drift free multiresolution encoder with several prediction loops. The prediction error of the low frequencies is reduced by 2.9 dB on the average for critical test signals compared to the reference method.