Standardized in 2014, multiview extension of high efficiency video coding (MV-HEVC) offers significantly better compression performance of up to 50% for multiview and 3D videos compared to multiple independent single view HEVC coding. However, the extreme high computational complexity of MV-HEVC demands significant optimization of the encoder. In this work, we propose a series of optimization techniques at various levels of abstraction: non-aggregation massively parallel motion estimation (ME) and disparity estimation (DE) for prediction units, fractional and bidirectional ME/DE, quantization parameter-based early termination of coding tree unit (CTU), and optimized resource-scheduled wave front parallel processing for CTU. When evaluated over three views for all available official multiview video coding test sequences, proposed optimization outperforms the anchor encoder by average factor of 5.4 at the cost of 4.4% bitrate (DBR) increase at no loss in PSNR, or alternatively a PSNR degradation of 0.12 dB at no change to the DBR.

中文翻译：

---

HEVC多视图编码的多级复杂度降低

与多独立单视图HEVC编码相比，高效视频编码（MV-HEVC）的多视图扩展在2014年标准化，可为多视图和3D视频提供高达50％的显着更好的压缩性能。但是，MV-HEVC的极高计算复杂度要求对编码器进行重大优化。在这项工作中，我们提出了各种抽象级别的一系列优化技术：用于预测单元的非聚集大规模并行运动估计（ME）和视差估计（DE），分数和双向ME / DE，基于量化参数的提前终止编码树单元（CTU）的优化，以及优化的CTU资源调度波前并行处理。在针对所有可用的官方多视图视频编码测试序列的三个视图进行评估时，