基于两阶段修复模型的虚拟视点合成研究

VIRTUAL VIEWPOINT SYNTHESIS BASED ON TWO-STAGE INPAINTING MODEL

  • 摘要: 在绘制虚拟视点图像后往往会出现空洞、重叠和伪影现象,为了提高虚拟视点图像的合成质量,提出基于两阶段修复模型的虚拟视点修复方法。采用双视点映射合成出虚拟视点图像,通过使用部分卷积的空洞填补网络进行空洞填补,通过图像优化网络去除图像中的重叠、伪影和部分卷积带来的局部色差。实验结果表明,修复后的虚拟视点图像在保证原有纹理信息的前提下,能准确地处理图像中的空洞、重叠、伪影等问题,与经典算法对比客观评价指标PSNR和SSIM分别高达31.878和0.9207,具有较高的绘图质量。

     

    Abstract: In order to improve the synthesis quality of virtual viewpoint image, a virtual viewpoint repair method based on two-stage repair model is proposed to solve the problem of holes, overlaps and artifacts after virtual views synthesizing. The virtual viewpoint image was synthesized by double viewpoint mapping, and the hole was filled by using the hole filling network of partial convolution. The overlap, artifact and local color difference caused by partial convolution in the image were removed through the image optimization network. The experimental results show that the restored virtual viewpoint image can accurately deal with the problems such as holes, overlaps and artifacts in the image on the premise of ensuring the original texture information. Compared with the classical algorithms, the objective evaluation indexes PSNR and SSIM reaches 31.878 and 0.9207 respectively, which has high drawing quality.

     

/

返回文章
返回