Research On Content Generation Method For The Naked-eye Three-dimensional Display

Recently, autostereoscopic displays receive extensive attention and make great progress. Along with the resolution of the naked-eye three-dimensional (3D) monitor and the number of viewpoints increasing gradually, the 3D display content generation faces new challenges in computing speed and memory storage. Limited to the physical size of cameras, film conditions, and transmission bandwidth, it is difficult to capture super multi-views directly for the naked-eye monitor. To transform images captured under existing conditions to images in stereo format for 3D display, it is necessary to restore the depth information and build virtual images for new viewing positions. The processing difficulty lies in stereo matching for depth estimation and virtual view rendering these irreversible problems. To process the large amount of calculating data from dense high-resolution views, it is necessary to design algorithms in parallel to synthesize stereo data accurately and efficiently.Here, to solve the above problems encountered in the content generation, we focus on improving the quality and synthesizing efficiency of images for the naked-eye 3D display. The main research and achievements of this paper are as follows,(1). An automatic parameter estimation method based on the degree of texture overlapping in accurate cost aggregation stereo matching is proposed.Estimation of the regularization parameter which strikes a balance between the spatial distance and color difference, is essential for successfully solving ill-posed image matching problems. Based on the cost filtering algorithm, a degree of texture overlapping (DTOL) is designed to simultaneously estimate the optimal regularization parameter and achieve accurate matching results. Experimental results demonstrate that the smoothing parameter is well estimated, and the disparity accuracy is improved to different kinds of stereo images in different illumination conditions. The method can be used to generate disparity sequences for different types of 3D videos automatically and efficiently.(2). An efficient spatiotemporal stereo matching method is proposed.The inter-frame information is added to the cost volume, which maintains continuity by one-dimensional iterative aggregations from horizontal, vertical and temporal directions. Compared with other spatiotemporal matching approaches, the algorithm has the advantage of fast calculation speed and avoids the mutual interference between inter-frames. Thus the matching accuracy and the jitter phenomenon between video frames are reduced. The continuity of stereo videos is maintained well.(3). An efficient multi-view video conversion structure in parallel and a predicted hole mapping algorithm are proposed.The efficient 3D video conversion structure based on GPU synthesizes images in 3D format directly from reference views and corresponding depth information without virtual views saving or interweaving processes. It leads to considerable reduction of computing time and memory footprint compared with traditional methods and facilitates a real-time 3D conversion system. In the 3D content generation by depth image based rendering (DIBR), a predicted hole mapping (PHM)algorithm is presented, in which the sparse sampling structure is designed for computing suitable mapping pixels to preserve the disparity continuity and enhance the sense of immersion. In experiments, the proposed PHM is evaluated and compared with some other methods in terms of PSNR and SSIM, and the result shows its advantages in the numbers. The average values of SSIM and PSNR are improved by 5.7% and 3.16% respectively.The method can operate on the 32-view display with 3840*2160 resolution in real time on GPU.(4). An adaptive joint view optimization algorithm is proposed and a three-to-dense view synthetization system is presented.The adaptive joint view optimization (AJVO) method enhances image edges by eliminating color-mixed pixels. Then visual artifacts in generated views are avoided fundamentally. With the preprocessing instead of post-processing on each virtual image, the computing time is reduced apparently,especially when the number of output views is great. Experimental results show that the generated dense views are well presented on the autostereoscopic display. Both the quality of virtual images and the continuity among viewpoints are improved by the system. Full-parallax stereo images can be obtained by the proposed ray searching method on images to reconstruct optical field for integral imaging 3D display accordingly.(5). A three-dimensional scene augmented reality method based on the depth information is proposed.Stereo matching, depth space transformation and scene fusion algorithms are presented to calibrate virtual cameras and real cameras under the condition without calibration board. Experiments show that the method improves the reality of fusion scenes, and realizes augmented reality on the naked-eye 3D display.