Fast HEVC Encoding with GPU Assisted Reference Picture Selection

Abstract

The emerging video coding standard, HEVC, provides a superior compression efficiency by exploiting flexible motion block partitioning and multiple reference pictures in typical settings. However, to obtain the compression potential of HEVC needs huge amount of computations to get the right motion vectors and associated reference picture indexes for each block, which may take most of the computational resources of encoding. In this paper, we tackle this problem by exploiting the computational capabilities provided by a GPU, to reduce the HEVC encoding time while preserving its high coding efficiency. We analyze the motion estimation module and divide the task into two parts. One favors GPU computing model, which involves lots of calculations without consideration of neighboring states. The other favor CPU computing model to consider the coding results of neighboring block, which can maintain HEVC’s high coding efficiency. In such a way of joint CPU-GPU encoding, we show that the encoding time can be significantly reduced. Comparing with the HEVC reference software, experimental results show that our proposed scheme achieves 34.4% encoding time reduction on average while the BD-rate increase is only about 2% for a typical lowdelay setting.