视点合成中基于深度的空洞修复  

Depth-based hole filling in view synthesis

作　　者：杜歆[1] 邹泷 

机构地区：[1]浙江大学信息与电子工程学系,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2015年第9期1616-1624,共9页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61271339);浙江省自然科学基金资助项目(LY12F01019)

摘　　要：为提高视点合成中空洞的修复质量,基于样本图像修复技术,提出利用深度信息进行修复的方法.通过深度图预处理、图像映射变换、合成深度图修复以及合成视图修复等步骤,实现对空洞区域的填充,得到虚拟视点视图.所提出的深度边缘单向膨胀预处理方法、先修复合成深度图再修复合成视图的策略以及匹配块的预编辑方法,可以提高最终合成视图的质量.实验结果表明:所提出方法在主观视觉对比、峰值信噪比(PSNR)以及运行耗时上优于现有方法,在图像边缘连续性的保持、前景和背景像素渗透现象的消除等方面具有较好的效果.An image inpainting method based on exemplar-based image restoration technology was proposed to improve the quality of hole filling in view synthesis by using depth information.This method fills the holes and provides a synthesized view through depth map preprocessing,image warping,depth inpainting and color image inpainting.The final synthesized view quality is improved by the depth edge＇s unidirectional dilation,the strategy that firstly inpaintes synthesized depth map and then the synthesized image,and the pre-editing of matched patch.The experimental results show that this method is superior to the existing methods in terms of subjective visual perception and objective index,such as peak signal to noise ratio（PSNR）and time consuming.The proposed method also has outstanding effects on maintaining the foreground edges and eliminating the infiltration from foreground to background.

关 键 词：视点合成 空洞修复 深度图修复 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]