Studies On Motion Estimation Techniques In Video Coding Standards

Video coding techniques have achieved great advances since stepping into the 21st century. The newly developed video coding standards, such as the H.264/AVC, have more than doubled the coding efficiency in major applications relative to the currently prevailing MPEG-2 video coding standard. Based on that, research efforts from both industry and academia have been made to investigate the next-generation video coding techniques in recent years, expecting further advances in new video-related applications.In those newly developed standards and those on-going research projects, motion estimation has played a very important role. It is one of the key technologies to reduce the redundancy within video signals and to improve the coding efficiency. However, the implementation of this technique will bring numerous block-matching operations, which introduces a heavy burden to the encoder in terms of computational complexity. Therefore, research on motion estimation is always a hot topic. After the standardization of a new video coding standard, such as the H.264/AVC, a great number of good motion estimation schemes are proposed based on it. These methods can achieve a very good trade-off between coding performance and the encoding speed. However, as the domain of video coding applications is expanding, originally designed motion estimation algorithms may have performance drops with a difference set-up of signal properties or encoding parameters.This thesis provides an intensive study on the design of motion estimation by researching the motion estimation related techniques. Based on that, new algorithms are proposed, in order to solve the performance instability problem when traditional methods are used in new environments. Firstly, according to theoretical analysis and experimental proves, the inefficient design of the original method is pointed out. Then an adaptive early termination threshold design—UMHS, is proposed, which can solve the performance instability of the original method with different video resolutions and quantization parameters. Secondly, the computational redundancy of the two-pass encoding procedure is summarized by analysis on implementation of the adaptive interpolation filtering technique. Moreover, a refined motion estimation strategy—pseudo one-pass ME (POME), is proposed to the adaptive interpolation filtering technique, which can greatly reduce the extra computational complexity when implementing this technique. Lastly, for multi-view video coding, the unique property of disparity estimation is analyzed from motion estimation. Reasons and principles for disparity and its distribution are concluded. Indicated by the global disparity direction, a new inter-view search range prediction method is proposed. This method can greatly reduce the computational complexity in disparity estimation by introducing an unsymmetrical search window design.