A Motion Attention Model Based Rate Control Algorithm for H.264/AVC

Rate control plays a very important role in video streaming applications. The currently adopted rate control scheme in the latest video coding standard H.264/AVC still results in a relatively large quality fluctuation especially for frames with higher motion activities. In this paper, we consider the characteristics of human visual system and incorporate a motion attention model to direct the rate control process. From the motion vector (MV) field at 4times4 block level, motion saliency maps are calculated on multi-scale MV fields and integrated to a whole motion saliency map highlighting the attention regions. In the frame and MB layer of rate control, the number of bits allocated to each frame and each MB is adjusted based on its motion saliency. More bits are allocated those frames and MBs with higher motion saliency to improve their qualities, and thus reduce the quality variation. Experimental results demonstrate that our algorithm can increase the average PSNR by 0.068 dB, and decreases the PSNR variation by 15.65%, with a negligible increase of computation less than 3%.