摘要
在分析UMHexagonS运动估计算法的基础上,根据中心偏移特性在起始搜索点预测中加入了原点预测;利用预测运动矢量的运动信息把当前块的运动类型区分为慢速运动、中速运动和快速运动,并采取相应的搜索策略;考虑到常见视频序列水平、垂直方向运动相对剧烈的特性,提出了用13点中菱形模板和8点多层次大菱形模板分别代替正方形模板和多层次六边形模板;在中菱形搜索和大菱形搜索的步骤后均加入提前终止判定,减少了对不重要搜索点的搜索。仿真结果表明,改进后的新算法在保证信噪比和编码码率基本不变的情况下,使得运动估计的时间减少了10%～23%,增强了编码的实时性。
On the basis of analysis UMHexagonS motion estimation algorithm, the original point prediction is added to the initial search-point prediction according to the center-offset characteristic. The motion type of the current block is divided into slow motion, medium-speed motion and fast motion by u- sing the motion information of the prediction motion vector, and then for different motion types use different search strategies. Considering the motion characteristics that horizontal and vertical direction relatively more intense of the common video sequences, this paper use 13 points diamond template to replace the square template, use 8 points multi-diamond template to replace the multi-hexagon template. Carry on early termination judgment behind the 13 points diamond search and 8 points multi-diamond search to reduce the search for less important points. Simulation result shows that the novel algo- rithm after improvement can reduce 10% - 23% of motion estimation time while keeping the SNR （Signal to Noise Ratio） and bit rate almost un- changed. The real-time characteristic is strengthened.
出处
《电视技术》
北大核心
2012年第5期15-17,25,共4页
Video Engineering
基金
国家自然科学基金项目(61007012)
浙江省科技创新团体项目(2010R50020)
作者简介
杨齐（1987-）。硕士生。主要研究方向为视频编解码技术。
李子印（1978-）。博士，硕士生导师。主要研究方向视频压缩与通信、图像处理技术。
