【摘要】 新一代视频编解码标准H.266/VVC在多个编码核心模块引入诸多新的技术，旨在提高视频编码的压缩效率，同时也带来复杂度显著增加。在块划分方面，在H.265/HEVC中四叉树划分（QT）的基础上增加了多叉树划分（MT），导致划分情况增多；此外帧间引入众多编码模式，使得确定每个CU的帧间最优模式的计算过程变得更加繁复，这些帧间预测技术极大增加了复杂度，阻碍了其在实际中的应用。因此提出了基于LightGBM的快速帧间块划分算法，通过分析常规的时空以及上下文特性，还引入多个新特征。通过选择大量相关性较强的特征并建多个LightGBM模型，并以此设计了全新的块划分流程，使得整体块划分过程显著缩短。实验结果表明，与H.266/VVC参考软件VTM10.0相比，平均降低33.98%计算复杂度，且BDBR仅增加了1.77%。更多还原

【Abstract】 The new generation video codec standard H.266/VVC introduces many new technologies in several core coding modules, aiming to improve the compression efficiency of video coding. But it also brings a significant increase in complexity. In the aspect of block partitioning, multi-tree partitioning(MT) is added on the basis of quadtree partitioning(QT) in H.265/HEVC, which leads to more partitioning cases. In addition, many coding modes are introduced between frames, which makes the calculation process of determining the optimal mode between frames of each CU more complicated. These interframe prediction techniques greatly increase the complexity and hinder its application in practice. Therefore, this paper proposes a fast interframe block partitioning algorithm based on LightGBM, and introduces several new features by analyzing the conventional spatio-temporal and contextual characteristics. By selecting a large number of highly relevant features and building multiple LightGBM models, a new block partitioning process is designed, which significantly shortened the whole block partitioning process. The experimental results show that compared with H.266/VVC reference software VTM10.0, the computational complexity is reduced by 33.98% on average, and the BDBR is increased by 1.77%.更多还原