A linear rate control model for better target buffer level tracking in H.264

Abstract

An improved rate control scheme with better buffer level tracking and more accurate mean absolute difference (MAD) prediction for H.264 video encoding is proposed in this work. Compared to the existing H.264 rate control scheme, the proposed scheme gives better buffer regulation with improved peak signal-to-noise ratio for basic unit level rate control. The bits required for encoding header information account for a significantly high percentage of the total bits, especially at lower bit rate and for low motion sequences. A linear relation between coded mean absolute difference and header bits is proposed for achieving effective rate control. The proposed scheme estimates the mean absolute difference adaptively from temporally colocated regions and INTER16x16 mode decision process. The estimated mean absolute difference is then used for calculating the quantization parameter for better target matching. A new linear rate control model is proposed, which works well especially for low motion sequences. The experimental results show that the proposed improvement in rate control algorithm significantly reduces the difference between actual and target buffer level while improving the quality of encoded video sequences.