Nowadays 3D films are becoming more and more popular due to their higher realism over the conventional two-dimensional (2D) ones, and 3D technique has been a trend of film industry. However, there are some special technical requirements for the choice of subject matter, shooting, editing, post-production to distribution and screening in 3D movies, which may lead to higher production cost and relatively longer production cycle to 2D ones. Although nearly one hundred 3D movies have come out, the shortage of 3D media sources is still serious, comparing with tremendous amount of 2D movies, 2D/3D video conversion technique provides an effective way to alleviate the problem, and has become a new research area in industry and academia recently. In this thesis, we focus on two aspects: the first is to design an effective 2D/3D conversion system; the second is to analyze some key techniques in the process of video conversion. The main contributions of this thesis include the following issues: (1) We make a thorough investigation on the emerging field of 2D/3D video conversion. Firstly, the various 2D/3D video conversion methods are classified and compared according to the relative motion between camera and scene. Then, several characteristics and difficulties of current researches are discussed. Finally, some promising future directions and implementation schemes are listed. On the basis, we further discuss the design of a video conversion system based on structure from motion (SFM) technique. Experimental results verify the validity of the key techniques proposed in this thesis, and also show that our video conversion system is of success for special 2D video clips. (2) We propose a new framework of piece-wise SFM, suitable for 2D/3D video conversion. SFM is an important research direction in computer vision, but its application in the 2D/3D video conversion is not mature. The new framework can recover space structures of sequential scenes from video streams and can provide depth cues for 3D video conversion. (3) Traditional depth normalization method cannot extract stable depth cues from video streams for 2D/3D conversion. To solve the problem, we propose a new depth normalization method based on the theory of parallax zones. Not only could this method realize stability in the process of depth extraction, but also enhance the stereoscopic feelings according to the characteristics of 3D displays. (4) We thoroughly study the depth image-based rendering (DIBR) ...
修改评论