New method for reducing GOP-boundary artifacts in wavelet-based video coding

The conventional motion-compensated temporal wavelet transform using the 5/3 filter produces GOP-boundary artifacts, i.e., a drop in picture quality at the boundaries of groups of pictures (GOP). A simple and efficient method is proposed in this paper to reduce the GOP-boundary artifacts and improve the overall performance of the transform. With this method, a group of pictures to be temporally transformed is extended by boundary repetition to one side and by symmetrical extension to the other side. The sub-sampling rule changes from one level to another during the multi-level transform. In addition, a non-uniform quantization scheme is employed. This method does not require any additional computation or memory. Experimental results of video coding with six test sequences show that this method outperforms the conventional method, providing a coding gain of 0.27 dB in terms of average PSNR and up to 1.95 dB in terms of minimum PSNR