A Perception-Based Rate Control Algorithm for H.264/AVC

Rate control plays an important role in H.264/AVC video coding standard. For the frame layer rate control, since the estimated values of quantization parameter (QP) of the initial frame vary from the image complexity, this paper firstly proposes an improved method for the initial QP estimation to make it more accurate. For the macro block (MB) layer rate control, an improved MB layer rate control algorithm based on human perception is also proposed in this paper. The proposed method classifies MBs by how visually attractive the MBs are and set perception-weights to different MBs. Then the encoder adjusts the QP according to the classified level and combines the perception-weights with structural similarity (SSIM) index to build up an improved rate-distortion optimization (RDO) model. Our experimental results show that the proposed algorithms outperform the original encoder x264 (Version 0.76.X) and significantly reduce the output bit-rate up to almost 50% without dropping down the visual quality of the video for low bit-rate applications.