Abstract: In free viewpoint video systems, obtaining high quality video on terminal display devices has always been a primary target for DIBR（Depth-Image Based Rendering） technology. The pixel interpolation is an integral part of DIBR, which greatly affects virtual view’s quality. This paper reveals the problems in the interpolation of standard rendering schemes and introduces a spatial weighting based synthesized method to optimize them. It takes into account the information in horizontal position and depth values of different posting point from referenced views. In the process of pixel interpolation, this method evaluates the influence of mapped pixels’ number in different areas on interpolation’s accuracy, removing part of the distorted pixels in synthesized images. Extensive experimental results show that the proposed method acquires satisfying effect both subjectively and objectively. Compared with the standard DIBR schemes, the method can improve PSNR and SSIM, respectively with 0.3 dB and 0.0013. Thus, the method can effectively restrain distortion in the process of interpolation.

Key words: synthesis, virtual view, Depth-Image Based Rendering（DIBR）, interpolation, spatial weighting

摘要： 在自由视点视频系统中，如何能在视频终端得到高质量的视频图像已成为基于深度图的绘制（DIBR）技术所研究的主要任务，其中虚拟视点像素插值是该技术中影响绘制质量的一个重要环节。针对虚拟视点绘制标准方案中存在的问题，提出了一种基于空间加权的像素插值算法。它是通过对多个投影像素点的深度值和水平方向绝对距离进行加权操作来实现像素插值的。在插值过程中，该算法考虑了不同区域投影像素点个数对像素插值准确性的影响，从而剔除了部分失真像素点，并且在图像输出前还分别对左、右参考虚拟视点进行了失真检测和矫正。实验结果表明，该算法改善了绘制的主、客观质量，其中，PSNR平均提高0.30 dB，SSIM平均提高0.001 3。因此，该算法可以有效地抑制像素插值过程引入的噪声，提高像素插值的精度。

关键词: 绘制, 虚拟视点, 基于深度图的绘制, 像素插值, 空间加权