Research And Optimization On Motion Estimation Technology In Scalable Video Coding

Scalable Video Coding (SVC) is an ongoing video coding standard as an extension of H.264/Advanced Video Coding (AVC). The goal of SVC is to provide scalability at the bitstream level with good compression efficiency by allowing combinations of scalable layers, including the spatial layer, temporal layer and Signal-to-Noise Ratio (SNR) layer. This enables a simple and flexible solution for transmission over heterogeneous networks, additionally providing adaptability for bandwidth variations and error conditions. However, an exhaustive search technique is proposed to select the best coding mode with the variable block-size motion estimation being employed. As a result, extremely high computational complexity is introduced, which seriously obstructs SVC from practical use. Therefore, it is much desirable to develop optimization algorithms to reduce the computational complexity of SVC without compromising the coding efficiency.In this paper,the basic knowledge of scalable video coding techniques and their application in H.264 scalable extensions are introduced. Motion Estimation (ME) techniques in H.264 video coding standard are analyzed in detail, including principle of ME, macroblock partitions, mode decision, R-D optimization, motion vector search algorithms and so on. And video quality measurement is introduced.Then, this paper analyzes the mode-distribution correlation between the base layer and enhancement layers in space layers and fine grain scalability(FGS) layers, and exploits the mode-distribution correlation to build fast mode decision algorithms which can reduce the computational complexity of SVC. Experimental results have demonstrated that the optimization algorithm can achieve a reduction of 50.34% encoding time on average, with a negligible average PSNR loss of 0.04dB and a 0.096% bit rate decrease.Also, in this paper, the distribution feather of vedio sequence is analyzed. It makes use of the information of temporal and spatial correlation in the encoded adjacent macroblocks, motion correlation between the temporal colacoted macroblock and current maroblock, judges beforehand the most possible modes then has a mode subset. The algorithm takes mode decision in the mode subset. Thereby the whole procedure of mode decision is simplified. The simulation results show that, compared to the original SVC reference software, this method can reduce the total encoding time by over 56.7% with a negligible average PSNR loss of 0.04dB and a 0.5% bit rate decrease.Finally, this paper analyzes the motion correlations in several macroblocks which are in the same object in vedio sequences, and proposes the concepts of motion energy and motion consistent characteristic. It designs the adaptive integer pixel and sub pixel motion vector search algorithm by judging the motion information. The experimental results show that the proposed algorithm can reduce encoder complexity significantly while coding quality is only negligible decreased.