基于SVM的MVC快速宏块模式选择算法

SVM-based Macroblock Mode Decision Algorithm for MVC

JMVC编码器采用全搜索算法在视点间和时间参考帧中遍历所有可能的预测模式选择最优,以获得最佳编码质量,在获得高压缩比的同时也带来非常高的编码复杂度,严重影响多视点视频的实际应用.针对此问题,利用JMVC多视点视频编码模型提出一种基于SVM的多视点视频编码快速宏块模式选择算法,利用SVM方法训练参考视点中提取到的相关信息,进行分级3层的SVM归类.对采用的3层分类策略进行讨论,并对特征参数的选取进行详细描述.实验结果表明,与JMVC全搜索算法相比,本文算法可在保持视频编码质量的同时节省大量编码时间.

The exhaustive full mode decision algorithm is applied to both temporal and inter-view prediction by joint multi-view video coding （JMVC） to achieve the best performance for multiview video coding （MVC）. Both the high compression ratio and the high complexity will seriously affect the practical use. A SVM-based macroblock mode decision algorithm for MVC using a JMVC platform is proposed in this paper. A three-stage SVM classifier is designed and the coded information between views is used as training data. In this paper, the applied three-stage classification strategy is discussed and the selected feature vectors are described in detail. Experiment results show that compared with the JMVC full mode algorithm, the encoding time can be remarkhaly saved by the proposed algorithm while maintaining nearly the same video quality.