Abstract
Motion estimation (ME) plays an important role in H.264 not only because it has extremely computational complexity, but also it affects the following block coding modes as well as the final coded bit-stream size. Therefore, studying on advance ME algorithms is one of the most efficient ways to improve the coding efficiency of a video codec. In thesis, we exhibit two ME schemes with software- and hardware-based implementations, respectively. Based on the studying on fast block-matching algorithms (FBMAs) and a priority list, the software-based ME scheme is started with a study on FBMAs, and then their efficiencies are illustrated in terms of algorithm checking points. A priority list is introduced to help with classification of motion content types of real world sequences. After doing statistical analyses on the proposed priority list and FBMAs, we propose a motion-content adaptive FBMA, which can adaptively switching searching strategies among three different FBMAs to maximize the coding efficiency under the considerations of motion-content variations. The latest H.264/AVC video coding standard adopts the variable block size (VBS) block partitions and multiple reference frames (MRF), which make the motion-compensation stage become extreme complicated. To save intermediate memory and maximum the hardware utilization, we propose an embedded merging scheme with a pipeline-based MRF extension. With this embedded design, only one copy of intermediate memory is required and fully utilization is expected after fulfilling the pipeline stages.