摘要
针对单视点绘制虚拟视点图像存在的空洞问题,提出一种基于深度引导空洞修复的虚拟视点绘制方法。核心思想是通过对多个映射像素点的深度信息和空间位置信息进行加权插值操作来实现细小的空洞填充,并且分区域对大型空洞填补。在像素点插值的过程中,该方法首先考虑映射像素点的可靠性,剔除不可靠的像素点,从而减少错误的映射像素点。在填补大型空洞中,继承基于样本图像修复算法的优点,并引入深度信息来辅助空洞填补,以较为准确的方式来修复图像。实验结果表明,无论是主观视觉效果,还是客观值都要比Criminisi算法要好。
A virtual view rendering method based on depth-guided hole filling is proposed to solve the problem of the existence of hole of single point virtual view image. The core idea is to realize the small hole filling by weight interpolation of the depth information and spatial position information of the multiple mapping pixel points, and fill the large holes in the each region. In the process of pixel interpolation, this method first considers the reliability of the mapping pixels and excludes the unreliable pixels, thereby reduces the error mapping pixels. In the process of filling large holes, this method inherits the merits of exemplar-based hole filling, and uses depth information to assist the operation of hole filling in a more accurate way. Experimental results show that the proposed method is better than Criminisi algorithm in both subjective visual effect and objective value.
出处
《计算机应用与软件》
2017年第6期162-168,共7页
Computer Applications and Software
基金
国家自然科学基金重点项目(U1301257)
关键词
虚拟视点
基于深度图的绘制
深度
空洞填补
像素插值
Virtual view Depth-image-based rendering （DIBR） Depth Hole filling Pixel interpolation
