基于空洞填补的虚拟视点绘制

Virtual Viewpoint Rendering Based on Hole Filling

在虚拟视点绘制过程中,参考视点中被前景物体遮挡的区域在虚拟视点中会变得可见,该区域由于缺少图像信息容易产生空洞。为此,提出一种基于背景识别的空洞填补算法。采用局部阈值分割与加权平均的结构张量,改进优先级计算的方法,利用深度图与最大化类间方差算法,给出新的源空间识别方法,以解决相对背景的识别问题。实验结果表明,与基于样本的图像修复算法及其改进算法相比,该算法可以较好地解决空洞问题,并完成虚拟视点绘制。

Hole problem arises in the virtual viewpoint rendering that the regions covered by the foreground objects in the reference view may be visible in the virtual view, which is caused by the lack of image information. Aiming at this problem, a new hole filling method based background recognition is proposed. In this method, local threshold segmentation and weighted average structure tensor are used to improve the calculation of the filling priority. An algorithm to recognize the source space is proposed, using the depth image and inter-class variance maximization algorithm, to solve the recognition problem of the relative background. Experimental results show that compared with the sample-based image in painting algorithm and its improved algorithms, this algorithm can effectively solve the hole problem and render virtual viewpoint well.