This paper presents a fast depth map encoding for 3D-High Efficiency Video Coding (3D-HEVC) based on static decision trees. We used data mining and machine learning to correlate the encoder context attributes, building the static decision trees. Each decision tree defines that a depth map Coding Unit (CU) must be or not be split into smaller blocks, considering the encoding context through the evaluation of the encoder attributes. Specialized decision trees for I-frames, P-frames and B-frames define the partitioning of $64\times 64$ , $32\times 32$ , and $16\times 16$ CUs. We trained the decision trees using data extracted from the 3D-HEVC Test Model considering all-intra and random-access configurations, and we evaluated the proposed approach considering the common test conditions. The experimental results demonstrated that this approach can halve the 3D-HEVC encoder computational effort with less than 0.24% of BD-rate increase on the average for all-intra configuration. When running on random-access configuration, our solution is able to reduce up to 58% the complete 3D-HEVC encoder computational effort with a BD-rate drop of only 0.13%. These results surpass all related works regarding computational effort reduction and BD-rate.

中文翻译：

---

使用机器学习的快速 3D-HEVC 深度图编码

本文提出了一种基于静态决策树的用于 3D 高效视频编码 (3D-HEVC) 的快速深度图编码。我们使用数据挖掘和机器学习来关联编码器上下文属性，构建静态决策树。每个决策树都定义了深度图编码单元 (CU) 是否必须拆分为更小的块，通过对编码器属性的评估来考虑编码上下文。I 帧、P 帧和 B 帧的专用决策树定义了 $64\times 64$ , $32\乘以 32$ ， 和 $16\乘以16$ CU。我们使用从 3D-HEVC 测试模型中提取的数据训练决策树，考虑到全帧内和随机访问配置，并在考虑常见测试条件的情况下评估了所提出的方法。实验结果表明，对于全帧内配置，这种方法可以将 3D-HEVC 编码器的计算工作量减半，而平均 BD 速率增加不到 0.24%。在随机访问配置下运行时，我们的解决方案能够将完整的 3D-HEVC 编码器计算工作量减少多达 58%，而 BD 速率仅下降 0.13%。这些结果超过了所有关于减少计算工作量和 BD-rate 的相关工作。