基于全景视频场景的VVC帧内快速划分算法

Fast Intra-Frame Partition Algorithm of VVC Based on Panoramic Video Scene

新一代视频编解码标准VVC使用了新颖的QTMT结构,相较上一代标准HEVC新增了二叉树划分和三叉树划分,而全景视频由于其视角原因包含了大量的数据,往往具有很大的分辨率,但这也导致全景视频编码时的时间复杂度骤升。为此,提出一种基于场景的VVC帧内快速划分算法,将待编码视频划分成不同的场景。此外使用SVM工具,为每个场景设置自己的SVM分类器,经过分类器分类后将冗余的划分模式剔除掉,从而优化编码器的时间复杂度。实验结果表明该优化算法以平均仅0.16%的编码性能损失换取了13.31%的编码时间节省,既保证了全景视频的编码效率,又提升了编码速度,具有优秀的表现。

This paper proposes a scene-based VVC intra fast partition algorithm,which divides the video to be encoded into different scenes.In addition,the SVM tool is used to set up its own SVM classifier for each scene.After classification by the classifier,redundant partition patterns are eliminated,thereby optimizing the time complexity of the encoder.The experimental results show that the optimization algorithm saves 13.31% of the coding time with an average coding performance loss of only 0.16%,which not only ensures the coding efficiency of panoramic video,but also improves the coding speed,and has excellent performance.