基于虚拟视图和低秩矩阵恢复的视点间预测

Inter-view Prediction Based on Virtual-view and Low-rank Matrix Recovery

新一代的多视点视频格式引入了深度图用于虚拟视图的合成。为了能充分利用深度信息,进一步提高多视点视频的压缩效率,本文提出一种基于虚拟视图和低秩矩阵恢复的视点间预测方法。首先利用深度图和相邻视点合成虚拟视图,当编码一个宏块时,利用虚拟视图中的对应块作为参考到相邻视图中寻找若干相似块,最后利用低秩矩阵恢复进行处理,以降低噪声。实验结果表明,该方法比JMVC(Joint Multi-view Video Coding)原有的视点间预测方法节省大约2%的码率。

The new generation of multi-view video introduced depth map for virtual view synthesis. In order to make full use of the depth information and improve the compression efficiency, this paper presents a prediction method based on the virtual view and low-rank matrix recovery. First, the paper generates the virtual view by the adjacent views and depth maps, and then sear- ches for several matching blocks from adjacent views using the corresponding block of virtual view as reference block. Finally, this paper uses low-rank matrix recovery to de-noise the prediction block. The experimental results show that it can save about 2% bit-rate than the original JMVC inter-view prediction.