一种新的立体显示中的视差估计与虚拟视点绘制算法

针对虚拟视图绘制技术中空洞和伪边缘等问题,提出了一种新的立体匹配算法以获得稠密视差图完成虚拟视点合成.此方法只需要两张视图,对其采用Mean shift算法就可以进行视图分割.有效的虚拟视点绘制是基于良好的视差图与好的视图插值法实现的.通过权值多窗口立体匹配法获得初始视差图,再利用交叉检验法对不可信匹配点和闭塞遮挡区域进行滤除,很好地解决了视差图中遮挡区域引起的误匹配问题.最后,对获得的稠密视差图进行视图插值和相应的噪声滤除完成了多点位置的虚拟视图绘制.针对Middlebury dataset提供的"Tsukuba"等测试图像进行相关实验,实验结果表明,该算法具有良好的视图绘制效果.

A stereo matching algorithm based on disparity estimation and virtual viewpoint rendering

Aiming at the problem of holes and false contour in virtual view rendering technology, a new stereo matching algorithm is presented in this paper to obtain dense disparity map to complete the virtual view synthesis.Only two images are required for this method, which are segmented using mean shift algorithm.A good view of the disparity map and an effective interpolation method are basis for successful virtual viewpoint drawing.The weighted multi-window stereo matching method is used to obtain the initial disparity map, and the cross checking is employed to filter false match points and occlusion parts.The obtained dense disparity map is then view interpolated and noise filtered to finish the multiple point virtual view drawing.Experimental results with test images like Tsukuba provided by Middlebury dataset show that the algorithm effectively improves the quality of virtual view images.