摘要
为了有效填补虚拟视点图像中的公共空洞,提出了一种基于逆向映射的空洞填补方法。首先利用深度图像绘制(DIBR)技术将左、右参考视点映射到虚拟视点位置,利用图像膨胀方法将映射的虚拟视图中的空洞区域进行扩大,以消除虚拟视点图像中的伪影瑕疵;然后,提取出膨胀后空洞区域的边界,并将其逆映射到原始的参考图像中,根据空洞与边界的相对位置,选取原始图像中相对位置上的像素来填充虚拟视图中的空洞区域;最后,将空洞填补之后的左、右视点映射的虚拟视图进行融合获得最终的虚拟视图。实验证明,本文方法有效解决了传统空洞填补方法容易将前景像素填充到背景区域的问题,能够获得较好的视觉观看效果和较高的客观峰值信噪比(PSNR)值。
In order to effectively fill the common holes in the virtual view imag e,we propose a novel approach of holes filling based on the backward 3D image warping method.In this method, first,the left reference view and right reference view are warped to the virtual view position with the t echnology of 3D image warping and the corresponding depth map information,respectively.To eliminate the ghosting art ifacts in the virtual view image, we enlarge the holes region in the warped virtual views with image dilating meth od.Second,the boundary of holes region is obtained and then backward warped to the original left and ri ght reference views with 3D image warping.Based on the relative position between holes and boundary,the c orresponding pixels in the reference views are chosen to fill the holes in the warped virtual views.Finally ,the holes filled warped virtual views are blended,and the final virtual image is obtained.Experimen tal results show that the proposed method can effectively solve the problem of filling foreground pixels into the b ackground areas with the traditional neighborhood interpolation method and image in-painting method.The proposed me thod also can obtain better subjective visual effects and achieve higher objective peak signal-to-noise ratio.
作者
王来花
齐苏敏
王玉田
倪建成
陈华
WANG Lai-hua 1, QI Su-min1 , WANG Yu-tian1 , NI Jian-cheng1 , CHEN Hua2(1. School of Software, Qufu Normal University, Qufu 273165, China ; 2. School of Information Science and Engi neering, Ningbo University, Ningbo 315211 ,Chin)
出处
《光电子.激光》
EI
CAS
CSCD
北大核心
2018年第3期318-324,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金青年基金(61601261)
山东省自然科学基金博士基金(ZR2016FB20)资助项目