The motion attention directed fast mode decision for spatial and CGS scalable video coding

The exhaustive search scheme for macroblock mode decision in the working draft of scalable video coding extension of H.264/AVC achieves theoretically optimal coding efficiency. However, this scheme also accompanies high computation complexity. Thus, for scalable video coding, how to reduce heavy computation load while there is minor bit rate increase and PSNR loss is a critical issue. In this paper, we propose a fast macroblock mode decision algorithm for scalable video coding. We consider the psychovisual issue and employing the correlations of macroblock partitions between base layer and enhancement layer. First, we adopt the original search scheme for macroblock mode decision in the base layer. The visually attended regions are identified based on the motion vectors in the base layer. Then, fast mode decision is applied to motion unattended regions in the spatial/CGS enhancement layer for time savings. It reduces mode candidates by exploiting the mode decisions in the previously encoded enhancement/base layer to predict the mode decisions in the current enhancement layer. Experiments conducted by JSVM8.10 exhibit that the proposed scheme saves 21.53% to 45.89% encoding time while there is a maximum of 1.7506% bit rate increase. It is noticed that there is a maximum of 0.1173 dB PSNR loss while the regions suffer from such loss are visually unattended.