加权多视图聚类方法应用于快速帧间模式决策的HEVC改进算法

Improved HEVC Algorithm With Weighted Multi-view Clustering Method Applied to Fast Inter Mode Decision

国际组织提出了新一代的视频编码标准:高效率视频编码标准.为了提高编码效率该标准引入大量新技术,随之不可避免的增加了计算复杂度.为进一步满足低延时视频传输的需求,本文提出将加权多视图聚类方法用于高效率视频编码帧间模式预测的想法,具体通过引入一种快速和高效的基于核的K均值多视图聚类算法,结合合并标签,运动矢量,跳过标签和率失真优化比率的特征选择.利用无监督机器学习视频帧间的时间相关性,达到减少帧间预测候选模式数量的目的.经过数学公式推导和实验证明,所提算法可以节省高达36.690%的编码时间,但只增加了0.278%的码率损耗.并通过与同类算法的横向比较,进一步证实所提算法的综合优势明显,达到预期目标.

International organizations proposed a new generation of the video coding standard,which called high efficiency video coding standard.In order to improve coding efficiency,the standard introduces a large number of new technologies,which inevitably increases computational complexity.So as to further meet the needs of low delay video transmission,this paper proposed the idea of using weighted multi-view clustering method for high efficiency video coding inter mode prediction.Specifically by introducing a fast and efficient kernel-based K-means multi-view clustering algorithm,combined with feature selection,such as Merge Flag,Motion Vector,SKIP Flag and RDO ratio.Using unsupervised machine learning the temporal correlation between video frames to reduce the number of candidate modes for inter-frame prediction.The mathematical formula derivation and experiments prove that the proposed algorithm can save up to36.690%of the coding time,and only add0.278%of the rate loss.And through the horizontal comparison with similar algorithms,it is further confirmed that the comprehensive advantages of the proposed algorithm are obvious.The expected goal was achieved.