递进填充线算法对DIBR虚拟图像的修复

发布时间：2018-05-14 14:20

本文选题：基于深度图的图像绘制 + 虚拟图像　；参考：《计算机辅助设计与图形学学报》2017年08期

【摘要】：在基于深度图的虚拟图像绘制(DIBR)中,前景物体的遮挡会造成虚拟图像中的空洞问题.针对Criminisi算法及模板匹配对空洞边界噪声敏感的问题,提出一种基于递进填充线的填补算法.首先根据视点变换的方向将空洞的背景部分边界扩大,以消除深度误差造成的空洞边界上的残余前景噪声;在空洞填充的遍历过程中,采用提出的递进填充线算法对每次更新后的剩余空洞提取基于视点方向的部分边界作为填充线,以限制被填充块的搜索范围,使被填充空洞块的择优顺序以背景纹理优先,避免了空洞边界上前景纹理的误填充.实验结果表明,该算法峰值信噪比比Criminisi算法平均高出1.617 3,比基于深度的模板匹配算法平均高出1.310 5,能更有效地抑制DIBR空洞纹理的错误生长,修复图像质量.
[Abstract]:In the virtual image rendering (DIBR) based on depth map, the occlusion of foreground objects will cause the void in the virtual image. A filling algorithm based on the progressive filling line is proposed to solve the problem of Criminisi algorithm and template matching sensitive to the noise in the hole boundary. In order to eliminate the residual foreground noise on the hole boundary caused by the depth error, in the traversal process of the hole filling, the proposed progressive filling line algorithm is used to extract the partial boundary based on the direction of the view as the filling line for the remaining holes after each update to limit the search range of the filled block and make the preferred CIS of the filled hollow block. In order to avoid the false filling of the foreground texture on the empty boundary, the algorithm avoids the false filling of the foreground texture on the hole boundary. The experimental results show that the peak signal to noise ratio of the algorithm is 1.6173 higher than the average of the Criminisi algorithm and 1.3105 higher than the depth based template matching algorithm. It can more effectively restrain the error growth of the DIBR hole texture and repair the image quality.

【作者单位】：四川师范大学基础教学学院;
【基金】：四川省教育厅项目(15ZB0043)
【分类号】：TP391.41

【参考文献】