Abstract: Based on the analysis of the characteristics of motion estimation in the spatial enhancement layer, a new algorithm is proposed. The algorithm estimates the prediction threshold of pixel SAD with the information from the base layer. It makes a decision of whether the motion estimation with the residual-prediction is terminated, so a lot of macroblock can avoid the undesirable motion estimation with the residual-prediction. And it makes use of the upsampled motion vectors to do the motion estimation with the residual-prediction if it is unavoidable. Simulation results indicate that the proposed algorithm can decrease the coding time of the motion estimation in the enhancement layer by 35％, and the computational complexity of encoder is significantly reduced.

Key words: scalabe video coding, inter-layer prediction, motion estimation