Multi-frame optimized quantization for high efficiency video coding

In video coding, the coding decisions made during the encoding of a reference frame impact the achievable rate distortion performance of all inter-predicted frames which refer to it. Most encoders operate frame-by-frame and do not specifically consider this inter-dependency. In this paper, a multi-frame transform coefficient optimization method for H.265/HEVC is proposed. The inter-frame dependencies are described using a linear signal model, such that the optimization problem can be cast in the form of an l1-regularized least squares problem. A simple functional relationship between the regularization parameter and the quantization paramter is empirically found. Bit rate savings are in the range of 7-10% for a simple IPPP... prediction structure and about 3% for a random access configuration.