Synthesis Image Quality Improvement Of Stereo Video Based On Virtual Viewpoint Synthesis

Three-dimensional(3D)video technology has been widely used in virtual reality,online live broadcast,three-dimensional television and other fields,which brings people a high-quality and realistic visual enjoyment.Virtual viewpoint synthesis technology is one of the core technology in the field of 3D video technology,which can synthesize video sequences shot at different positions into video images observed at other viewpoint positions,and then construct a multi-view 3D video.However,the current 3D video still has visual flaws,which will seriously reduce the quality and perceived comfort of the 3D video and affect the user experience.There are still many rendering problems in the composite video image sequence output by the virtual view synthesis method.These problems include overlapping problems,artifacts,and holes.In order to solve these rendering problems,this dissertation proposes a method to improve the quality of the existing 3D video virtual viewpoint rendering methods.The main contributions are as follows:1)First,conduct a detailed study and elaboration on the virtual viewpoint synthesis process in 3D video technology,and analyze the possible rendering defects;and propose an improved method for the virtual viewpoint synthesis process for RGB-D images.This method preprocesses the input depth video firstly,including noise removal and edge smoothing process;then by comparing and judging the depth values of different pixels in the rendering process,the problem of image projection overlap in the rendering process is solved.Aiming at the problem of artifacts in the rendered image,the method of hole diffusion is used to cover the foreground pixels incorrectly projected to the background area as the hole area for processing.Aiming at the hole problem in the rendering effect,an iterative algorithm based on the pixels around the image hole is used to fill holes.The experimental results show that the method proposed in this paper has a significant improvement in subjective quality.The PSNR is improved by about 0.76dB and 0.5dB compared with the VSRS method and the improved Criminisi method,respectively,and it is also improved by about 1%on SSIM than the two rendering methods.2)Propose an improved method to draw virtual view video sequences through multi-view video sequences.First divides the input image using the K-means method,and then draws the divided images separately and merges the virtual viewpoint drawing method.At the same time,a hole filling method based on the superposition of series is proposed.This filling method classifies the pixels around the hole pixels,and records and calculates the values of these pixels to obtain the final filling result.Experimental results show that the subjective and objective quality of the virtual viewpoint video sequence method proposed in this paper is higher than other rendering methods.