基于色彩比例和时域相关的置信度传播立体视频匹配算法

提出一种改进的立体视频匹配算法,首先通过马尔科夫随机场建立视差图模型,然后利用亮度和色彩比例求置信度,最后根据时域相关性局部更新视差值得到视差序列。以卡耐基梅隆大学提供的真实场景视频为测试序列,实验结果表明:基于亮度和色彩比例比单纯使用亮度得到的匹配更准确,根据时域相关性求视差序列比用逐帧置信度传播求视差序列减少了83.3%以上的迭代计算时间,为立体视频的压缩传输、任意视点视频合成提供了高效的视差序列。

Stereo video matching with color proportion and temporal correlation for belief propagation

An improved stereo video matching method was proposed.First,Markov random field model was used to build the stereo disparity model.Then the intensity and color proportion were employed to compute the belief.Finally the disparity sequence was computed by temporal correlation refreshed disparity value locally.Real scene provided by Carnegie Mellon University was tested.Experimental results show that the matching result based on luminance and color proportion is more accurate than the one based on luminance only.Using the temporal correlation matching method the computing iteration time could be reduced by 83.3%,which can provide high efficient disparity sequence for stereo video compression and free view point rendering.