| With the development of video technology,the traditional 2D video is developing towards the direction of free viewpoint video.The free viewpoint video allows the viewer to freely choose the viewing point,which can give the viewer three-dimensional experience and improve the immersive experience of the audience.But because of the limitation of transmission bandwidth in free viewpoint video system,the receiver of free viewpoint video system needs to use virtual view synthesis technology to draw the virtual image of the target viewpoint.Virtual viewpoint rendering technology based on depth image is the key technology in free viewpoint video,which has the advantages of fast speed and saving bandwidth,but the rendered image still has some problems such as voids,overlaps,artifacts,cracks and so on.In order to improve the quality of the virtual image of the target viewpoint,it is necessary to solve the above problems.In order to solve the problem of voids and overlaps in the virtual viewpoint image before double-view image fusion,the idea of layering is used to segment the foreground and background of the virtual viewpoint image after rendering.After the segmentation,the left and right virtual viewpoint images are respectively fused by the distance based fusion method,and the foreground and background images are obtained.This method can solve the overlaps problem while using the fused background image to guide the repair of the voids.Experimental results show that the PSNR value of this method is 1.75 d B higher than that of classic Criminisi algorithm,which can effectively improve the quality of rendering images.In order to solve the problem of voids and artifacts in virtual viewpoint image after fusion of two views,background information is used for post-processing.Firstly,the artifact regions in the left and right virtual viewpoint images are located according to the viewpoint movement direction,and the foreground artifacts are erased by the background information of complementary virtual viewpoints.Secondly,The reference viewpoint image is introduced into the voids processing process.Based on the redundancy of information between reference images of different viewpoints,the depth information of reference images of different viewpoints was used for fusion.Linear interpolation method is used to remove the remaining foreground pixels in the fused image,and then forward mapping is carried out to get the background image at the virtual viewpoint to deal with the voids problem.Finally,Compared with the virtual viewpoint rendering algorithm based on image segmentation,the PSNR value is improved by 0.636 d B.It improves the quality of the rendering images. |
PDF Full Text Request