对3D Warping分解的高质量虚拟视点绘制方法

High-Quality Virtual Viewpoint Rendering for 3D Warping

基于深度图像的虚拟视点绘制(DIBR)作为自由视点视频应用中的核心问题,由其获得的任意视频的质量和速度对于自由视点视频的发展至关重要。为解决经典DIBR方法存在的重叠、空洞、伪影,以及由于重采样导致的细小裂纹问题以提高虚拟视点图像质量,提出了一种基于改进的3D Warping过程的虚拟视点绘制方法。该方法从整个3D Warping过程出发,将3D Warping过程分解成两步而无需进行三维建模,并分别在两步中提出了改进的方法。在第一步中,通过在变换矩阵中引入可调的系数对三维参数进行修改;在第二步中,主要针对由于重采样导致的细小的裂纹问题,提出了自适应一投多算法,在兼顾时间复杂度的同时改善了裂纹问题。实验结果表明,无论从主观质量还是客观评价标准来看,该方法都能够显著提高虚拟视点绘制的图像质量。

DIBR is the core problem in the application of free view video.The quality and speed of arbitrary video is very important for the development of free view video.To solve the problem of overlapping,hollow,artifacts in the classical DIBR method,and the small cracks problem caused by resampling,this paper presents a virtual view rendering method based on improved 3D mapping process to improve the quality of virtual viewpoint image.This method starts from the whole 3D mapping process and does not need 3D modeling,but decomposes the 3D mapping process into two steps,and proposes an improved method in two steps.Firstly,the three-dimensional parameters are modified by introducing the adjustable coefficients into the transformation matrix.Then,this paper proposes an adaptive algorithm for a vote more for the small crack problem caused by resampling taking into account the complexity of the time while improving the crack problem.The experimental results show that the method can significantly improve the image quality of virtual view rendering,both from subjective quality and objective evaluation criteria.