On re-composition of motion compensated macroblocks for DCT-based video transcoding

Abstract

To achieve portability between different kinds of encoding formats and network environments, heterogeneous video transcoding becomes a key technique for reducing the bitrate of a previously compressed video signal. A frame-skipping transcoder is often used to avoid an unacceptable picture quality when high transcoding ratio is required. Due to high computational complexity and quality degradation introduced by conventional frame-skipping transcoders, a DCT-based video frame-skipping transcoder has been proposed recently. However, the transcoding process of the motion compensated macroblocks in the DCT domain becomes the bottleneck since IDCT and DCT processes are required. In this paper, we propose a new architecture of the frame-skipping transcoder to reduce the computational complexity of motion compensated macroblocks in the frame-skipping process. The new architecture transcodes the dominant region of a motion compensated macroblock in the DCT domain by making use of the DCT coefficients of the incoming bistream and some pre-computed shift operators. By using a shifted version of the dominant vector, the re-encoding error introduced in the dominant region can be avoided. On the other hand, an adaptive transcoding architecture to transcode the boundary regions of MC marcoblocks and a way to perform error compensation are proposed. This architecture can further speed up the transcoding process of the motion compensated macroblocks. Half pixel accuracy related to our proposed frame skipping transcoder is also addressed. Experimental results show that, as compared to the conventional or DCT-based transocders, the new architecture is more robust to noise, gives rise to fewer requantization errors, and requires simple computational complexity.