We present a novel video matting method for extracting multi objects’ matte information in a video simultaneously. Recently, video matting technology is received abundant interests in computer graphics community. With the very recent video matting, however, we can’t deal with multi objects because of occlusion problem between multi objects. For solving this occlusion problem, we adopt a full 3d segmentation approach. The key idea is that we consider a video space as a 2D scatter plot not 3D cube. By using this approach, we can solve three problems for multi object video matting. First, we solved occlusion problem between multi objects naturally. Second, for speed-up of full 3d segmentation, we can make video data more efficiently using mean-shift segmentation algorithm. Third, we can apply feature-based object tracking algorithm like SIFT (Scale-Invariant Feature Transform) for each pixel. This process makes our proposed method more robust to dynamic scene. In addition, by reducing the video dimension from 3D to 2D, we can achieve more accurate segmentation results.