基于多分支网络的深度图帧内编码单元快速划分算法

3维高效视频编码(3D-HEVC)标准是最新的3维(3D)视频编码标准，但由于其引入深度图编码技术导致编码复杂度大幅增加。其中，深度图帧内编码单元(CU)的四叉树划分占3D-HEVC编码复杂度的90%以上。对此，在3D-HEVC深度图帧内编码模式下，针对CU四叉树划分复杂度高的问题，该文提出一种基于深度学习的CU划分结构快速预测方案。首先，构建学习深度图CU划分结构信息的数据集；其次，搭建预测CU划分结构的多分支卷积神经网络(MB-CNN)模型，并利用构建的数据集训练MB-CNN模型；最后，将MB-CNN模型嵌入3D-HEVC的测试平台，通过直接预测深度图帧内编码模式下CU的划分结构来降低CU划分复杂度。与标准算法相比，编码复杂度平均降低了37.4%。实验结果表明，在不影响合成视点质量的前提下，该文所提算法有效地降低了3D-HEVC的编码复杂度。

Fast Partition Algorithm in Depth Map Intra-frame Coding Unit Based on Multi-branch Network

Three Dimensional-High Efficiency Video Coding (3D-HEVC) standard is the latest Three-Dimensional (3D) video coding standard, but the coding complexity increases greatly due to the introduction of depth map coding technology. Among them, the quad-tree partition of depth map intra-frame Coding Unit (CU) accounts for more than 90% of the coding complexity in 3D-HEVC. Therefore, for the intra-frame coding of depth map in 3D-HEVC, considering the high complexity of CU quad-tree partition, a fast prediction scheme of CU partition structure based on deep learning is proposed. Firstly, the dataset of CU partition structure information for learning depth map is constructed. Secondly, a Multi-Branch Convolutional Neural Network (MB-CNN) model for predicting the CU partition structure is built. Then, the MB-CNN model is trained by using the built dataset. Finally, the MB-CNN model is embedded into the 3D-HEVC test platform, which reduces greatly the complexity of CU partition by predicting the partition structure of CU in depth map intra-frame coding. Experimental results show that the proposed algorithm reduces effectively the coding complexity of 3D-HEVC without significant synthesized view quality distortion. Specifically, compared to the standard method, the coding complexity on the standard test sequence is reduced by 37.4%.