| In recent years,3D film attracts countless fans because of its vivid binocular vision,but traditional 3D video costs a lot and its time-consuming shortcoming limits its promotion.However,we can get the 3D videos quickly and low costly by depth-image-based rendering,it is regarded as the preferred method that can be used to replace traditional 3D video production.However,there will be many holes in the virtual view,which influences the quality of synthesized view seriously and at the same time relieves the visual fatigue.Filling the holes of high image quality is a key issue in DIBR.This dissertation aims at filling the holes with high image quality in the synthesized view of the DIBR system and promoting the development of DIBR technology in 3D video production.This dissertation discusses hole generation mechanism in two “reference image + depth map”,and an modified camera setup method that does not generate holes is proposed.The characteristics of the foreground and background pixels along the hole are analyzed in one “reference image + depth map”,and a superpixel-based hole-filling algorithm is proposed.In order to overcome the shortcomings of the superpixel-based hole-filling algorithm,an improved hole-filling based on inpainting is presented.The main research results of this dissertation are summarized as follows.(1)Based on the double-sided DIBR technique,a camera setup method that does not generate holes is proposed.The hole generation mechanism of two scenes is analyzed,and it can be concluded as follow: in “foreground-background-foreground” scene,regardless of how the camera parameters are set,the resulting virtual view has no holes;in “background-foreground-foreground” scene,you can get the virtual view with no holes by setting the camera baseline.The next step is to deduce the camera setup method under the situation of the unparallel camera setup.(2)Because background pixels are used to fill the hole,a superpixel based hole-filling algorithm is proposed in this dissertation.This algorithm uses SLIC-superpixel algorithm to segment the virtual view firstly,and then calculates the priority along the unilateral edge of the virtual view,in the end,the super-pixel segmented is regarded as the matching patch to fill the patch with the highest priority.The experimental results show that the proposed algorithm can distinguish foreground and background accurately,and at the same time it can get good filling results for the images of simple texture,but the artifacts are generated when both sides of the hole are foreground.(3)An improved hole-filling based on inpainting is proposed to solve the problems which can not be solved by the superpixel-based hole-filling algorithm.The holes with foreground on its both sides are filled by the hole-filling based on disparity map firstly,and the rest holes are filled by the improved hole-filling based on inpainting.The steps are as follows: at first,the position of the boundary hole are labeled by a flag,and then uses the improved computation method of priority to calculate the priority along the unilateral edge of the holes.Finally,the “squeeze” method is used to determine the searching range of the matching patch.The experimental results show that the proposed algorithm can fill the hole with foreground on its both sides and the boundary hole in different zero-parallax setting plane.At the same time,the proposed algorithm can restore the texture of the virtual view well,also it has certain advantages in time complexity.In this paper,different hole-filling methods of DIBR system in various application scenarios are proposed.The research results will provide a new method of practical applications of glasses-free 3D display technology. |
PDF Full Text Request