R-D Optimized Frame-Layer Bit Allocation for H.264

The frame-layer bit allocation problem for H.264 is studied in this work. We first impose a GOP (Group of Pictures) structure on the underlying H.264 video, and then develop a two-pass rate control algorithm using the Lagrange optimization framework. The proposed algorithm estimates the rate-distortion (R-D) information of all frames in a GOP via pre-encoding in the first pass. The target number of bits of each frame is determined by searching the Lagrange multiplier that meets the bit budget for the GOP with the estimated R-D information. The proposed algorithm is useful for non-conversational applications such as video streaming where video sources are encoded off-line. Experimental results demonstrate that the proposed algorithm achieves substantially higher picture quality than that used in the H.264 JM8.1a reference code.