Study On The Prediction Method Of Disparity Based On The Analysis Of Multi-view Decoding Stream

With the rapid development of science and technology, video and image are moreand more important in modern work and life of people, which have improved the workefficiency and quality of life. At the same time, the capture, compression and displaytechnology of video images are in rapid development. Stereoscopic vision technologieswhich can bring viewers more strong sense of reality have attracted increasing attentionof people. Currently, stereoscopic vision technologies have been widely used in teleme-dicine, virtual reality, films, cartoons and other fields, and were welcomed by people.However, comparing with2D video, a prominent problem of stereoscopic video isthe aggravation of visual fatigue. For traditional2D video, watching for a long time canmake people feel the sense of fatigue, while such sense of fatigue becomes worse afterwatching the stereoscopic video. Many viewers will suffer from the symptoms such aseye pain, dizziness, and blurred vision after watching stereoscopic video whichdepressed the interests of people watching stereoscopic video. Visual fatigue hasbecome a major obstacle for the further popularization of stereoscopic video.The factors which lead to visual fatigue of stereoscopic video mainly include: theexternal environmental factors, physiological factors of viewers and visualpsychological factors. The external environmental factors include mainly the playingenvironment and the display situation of stereoscopic video display equipment, whichcan be eliminated by optimizing the playing environment and display performance ofthe equipment. Visual physiological factors mainly refer to the factors which areinconsistency with the physiological characteristics such as vertical disparity and theinconsistency between convergence and focus adjustment. These factors need thescientific and systematic research to eliminate. At present, the inconsistency betweenconvergence and focus adjustment has been explored widely and deeply and has madegreat breakthroughs. However, the study on vertical disparity is relatively few.In real life, when people watching the same scene, two virtual images will beformed respectively in both eyes. According to the difference of these two images,people can acquire the3D information of the scene. When we use cameras to obtainstereoscopic video, it is just the simulation of such process, and the difference isreferred to as disparity. Disparity vector consists of horizontal disparity and verticaldisparity. The horizontal disparity, which is derived from the horizontal placements ofthe two cameras, is an important basis for3D reconstruction. And human eyes have hadthe adaptability to the horizontal disparity. Vertical disparity is caused by the placements and techniques of cameras. It makes no contribute to the3D reconstruction,and is not adapted to human beings. Furthermore, it will cause visual fatigue and affectstereoscopic video quality. Therefore it is necessary to be eliminated.Vertical disparity elimination algorithm consumes too much time, and may mistakehorizontal disparity. Therefore vertical disparity should be predicted, so that thedisparity elimination algorithm can be called appropriately to save time consuming orthe change of horizontal disparity. At present, prediction methods of disparity aremostly concentrated on the encoding end and used in fast encoding algorithms. Inpractical applications, vertical disparity elimination module is called at the decoding end,so the prediction of disparity should also be implemented at the decoding end. In thisthesis, disparity vectors of JMVC are predicted by using the decoding stream firstly, andthen the study on this disparity and the actual disparity of each frame is carried on, andthe relationship of these two disparities is established at last.The algorithms in the thesis are implemented by C++and OpenCV on the platformof Visual Studio2008. The multi-view video sequences ballet, ballroom, exit, crowdand vassar are used as the test video sequences. Experimental results show that theproposed model can predict the disparity of each frame in the multi-view videoaccurately.