Three-dimensional Display Information Reconstruction And Evaluation Methods

The world we live in is a real, three-dimensional (3D) space. From ancientpainting recording image around or in mind, to the birth of a camera and videotape recorder, people never stopped pursuing more vivid and lively record ofthe world. When "Avatar" shock up the notion of traditional2-D images,3-Ddisplay became a household name and a new popular pursuit in the displayarea.After enjoying the stunning visual shock of3-D images at the beginningof the21st century, people are eager to get more. However, large scale use of3-D display is still a challenge by now. Scarcity of3-D film sources is the mostimportant reason. The high-precision3-D camera can shoot any scene and usedfor3-D film and television production, but it is hard to widely spread due tohigh cost and long production cycle. At the same time, facing abundant andeasy making2-D images and videos, people also want to relive the samecontent under the3D shock effect. Therefore, the technology of3-Dinformation reconstruction came out.The3-D display information reconstruction technology, i.e.2D-3Dconversion, is to convert2-D images or videos to acceptable3-D display ofimages and video through software algorithm. In the conversion process, thedepth of the image sequence (i.e."depth map") will be used. Depth map isinformation gap between2-D display and3-D display. When adding depth map,a2-D image will be able to convert to3-D image by using renderingtechnology based on depth (DIBR-depth image based rendering).Theoretically,this is a practical approach to breakthrough the bottleneck of3D industrydevelopment–lack of sources.Related research started since the very beginning of21th century. Thereare also many encouraging research progress in this field. This paper, inChapter1, reviews the previous research of3-D technology, then introducesbackground knowledge of the research methods in our work, included basicprinciple of human eyes vision, principle and classification of3D displaytechnology. There are single eye depth perception and binocular parallax depth perception. We make a detailed analysis from three aspects, glassed typed3-Ddisplay, naked eyes typed3-D display and generalized3-D display；In Chapter2, this paper summarizes current research status of3-D displayinformation reconstruction and industry problems at this stage；meanwhile,points out the important status and realistic significance of3-D displayinformation reconstruction techniques in whole3-D video capture process.This chapter introduces concept of depth map, describes3-D image renderingtechniques based on depth map (DI BR) in detail, and analyzes current videoprocessing tools and research status in2D-3D conversion, including imageedge detection algorithm, image segmentation algorithms, pattern recognitionand video tracking.Following the logic of depth map based3-D rendering, Chapter3describeour exploration of a new2D-3D conversion system. The first step ispre-processing, including2-D image acquisition and bilateral filtering.Experiments show that bilateral filtering method we use in this paper for2-Dpreprocessing effectively lowered the error rate in2D-3D conversion. Afterthat, we classify objects in the image which has been pro-processed usingk-means clustering algorithm, and then image which foreground andbackground is divided and waiting for assigning depth value would be gotten.In order to assign depth value to the image more accurately, we propose layoutmodel of image which has been processed using k-means clustering algorithm.Experiment results shows that the method we proposed has explored anavailable2D to3D method, and laid the groundwork for future research.We also propose a semiautomatic, interoperability system through whichkey information of3D video from2D video can be captured. In the system,automatic generation method of non-key frames depth map base on objecttracking algorithm has been proposed. The system can be used to generatedepth information which generation of3D video required. First, get key framefrom video, do image segmentation based lazy snapping algorithm on keyframe, then label main object in key frame interactively, depth value will begiven after that. For non-key frames, according to depth map of key frames, weuse generation method of non-key frames depth map base on object trackingalgorithm which we proposed to generate depth map of non-key frame automatically. Finally, depth image sequence which corresponds to2D videosequence has been generated. We propose2D to3D system framework at theend of this chapter. The experimental results show that the generation methodof non-key frames depth map base on object tracking algorithm which wepropose can get correct result, depth map generation using this algorithm hasgood performance.While making deep research of2D to3D algorithms and system,considering3D display industry is still at the stage of promotion, we proposecomprehensive method of measurement for the image quality of3DTV. In viewof actual needs of this industrial development stage, we propose measurementconditions and measurement method for quality of flat panel stereo TV and flatpanel stereo terminal in the type of glassed3D display. We makecomprehensive research and rule on image quality of3DTV measurementrequirement. Include following aspects, working conditions, environmentalpresentation, power, stable duration, measurement site, measurement position,format of test signal, measurement points and measurement points,measurement instruments, standard test status. And we raise reference standardmeasurement methods. Research results have formed the industry standardtext.In addition, based on the concept of measurement uncertainty, this paperstudies the measurement uncertainty of set-top box’s electrical level test byusing SFU (Single Family Unit)-one kind of broadcast test system instrumentwhich is widely considered can generate more-or-less standard video signals.We propose evaluation methods aim at A-class and B-class measurementuncertainty respective. The evaluation methods and the experimental result canbe applied to evaluate image quality of3-D television.At last, we combined with the actual industry application experience,designs a family network system architecture and reference model which ismore adaptable in real family network construction scenarios in demanding of3-D display and high definition interactions. We also proposed a networkconvergence solution based on indoor Power-Line Carrier (PLC) network,which is already in approval process for national standards.