一种融合了信念传播思想的背景分割优化方法

An optimized background segmentation method combining belief propagation thought

背景分割问题是目标检测、目标跟踪、目标识别和目标理解等一系列运动分析的首要步骤。为了解决目前分割结果中出现的"空洞"现象,提出了将改进的Codebook背景建模方法和置信传播思想相融合的优化背景分割方法。首先,利用改进了匹配条件的Codebook背景建模算法对场景进行建模,减少边缘区域和场景较暗区域的噪声,得到初始的分割结果;然后,基于置信传播思想,通过消息传递,加强像素之间的关联,平滑初始背景分割结果中出现的"空洞"现象,达到优化的目的;最后,通过举重训练图像分别对算法的有效性进行检测。结果表明,证明优化分割方法可以很好地填充Codebook背景建模方法的分割结果中出现的"空洞"现象。

Although there are many background segmentation methods, relationships among pixels are often ignored in these methods, resulting in the ＂hole＂ in the extracted object. An approach combining codebook scheme and belief propagation scheme to optimize the segmentation object has been proposed. Firstly, the background scene is modeled using codebook scheme with im- proved matching conditions and the initial object is extracted. Then the optimal belial propagation scheme is used which couId fill the hole in the initial segmentation. Finally, the optimized approach is tested using practical video of weight-lifting. For the prac- tical video, it could obtain the initial segmentations and the optimization. These results show that the optimized approach can ob- tain more complete segmentation.