Study And Implementation Of Digital Video Non-Linear Editing

In the recent years, digital video products have penetrated various fields of daily life rapidly, and influence people's production and life strongly. Digital TV is substituting analog TV rapidly. A digital TV system includes digital video program producing and editing,digital signal transmission and digital signal reception. Furthermore, it can provide consumers better video and audio performance than analog TV does.NLE (Non-linear editing) is widely used in digital TV postproduction. A powerful non-linear editing system can support animation production, video clip editing, scene composition,special effect and subtitle integration. On the other hand, with the rapid development of the digital technology and the deployment of digital TV courses, and moreover, with MPEG-2 being adopted in popular digital TV standard such as DVB, ATSC and ISDB, their production and editing technologies appear totally different form the traditional ones. Therefore, NLE system should be able to deal with MPEG-2 programs.Recently, there are two typical NLE systems for MPEG-2 video editing, one is based on spatial domain and the other is based on the compressed domain. The main work and achievements of the author on nonlinear editing based on these two systems in recent years are systematically presented in this paper.In chapter 1, the basic theories and technologies related to NLE are introduced. The author classifies the NLE system based on different applications. Finally, new technologies and latest developments on NLE are also introduced.In chapter 2, the development and implementation of NLE software system based on windows is summarized. Based on the NLE software ayatem, the author designed and implemented lots of video special effects, such as regular geometric transformation, scaling, color adjusting, fade, dissolve and wipe. Further more, some algorithms for optimizing the performance of special effect are also proposed. Finnaly, the author design and implemented the real-time special effect based on the HD video board.In chapter 3, fast implementations of NLE special effects based on spatial domain are discussed. Firstly, the author analyzes the computation complexity within the NLE loop and find that the full-scale full-search ME in re-compressing process costs most of the time. Then, motion vector refinement algorithms for fast gradual transition special effects are proposed, which reduces the computation complexity by reusing the motion vector information extracted from the original MPEG-2 sequences. Experimental results show that the proposed adaptive motion vector refinement (AMVR) algorithms achieve comparable quality of the reconstructed edited video, higher PSNR with the lower computational complexity, in comparison with the conventional motion vector refinement algorithms.In chapter 4, based on the spatial domain NLE, rate control strategies during gradual transition special effect are proposed. The influences of bit allocation and GOP size in the gradual transition period are evaluated. Further more, by studying the TM5 rate control strategy in detail, the mechanism of accurate rate control and rational bit allocation is analyzed. And hence rate control strategy for gradual transition special effect is proposed. Results show that the rate control strategy can improve the video performance in the gradual scene transition period compared to TM5.In chapter 5, fast implementations of NLE special effects based on compressed domain are discussed. Firstly, the algorithms of image processing in compressed domain are presented, by which several special effects are realized such as scaling, geometric transformation and mirror image. Then MC-DCT is introduced to obtain the DCT domain frames from MPEG-2 sequences. Finally, novel algorithms of special effect in compressed domain for MPEG-2 sequences without full frame decompression and motion estimation are proposed. Results show that with the reduction of computational complexity, the quality of edited video in compressed domain is still close to the quality of the edited video in uncompressed domain at the same bit rate.