为了提升H.264编码器中运动估计UMHexagonS(unsymmetricalcrosshexagongridsearch)算法的效率,在深入研究算法的基础上对其做了改进。引入了一种新的早期结束搜索阈值,根据初始预测向量偏置性对粗搜索阶段的混合模板进行了区域划分和改进,针对不同大小的宏块应用不同的搜索模板,以上优化降低了运动估计算法的复杂度,极大的节省了运动估计时间。通过不同运动类型的视频序列测试实验表明,改进的算法与原算法相比,运动估计时间平均节省了17.93%,同时保持相近的码率和图像质量,总体提高了编码器的性能。
In order to increase the H. 264 encoder efficiency, three optimization methods for the UMHexagons based on deep research of the algorithm are proposed. Firstly, a new early termination (ET) threshold is introduced. Secondly, the hybrid search pattern in the coarse search step is partitioned and improved according to the direction characteristic of the motion vector, then various search patterns are used for different macro blocks in the next search step. All of the optimization not only reduce the complexity of the original algorithm but also save motion estimatiom time largely. The experiments with some standard video sequences of different types show that the improved algorithm can decrease approximately 17.93 % motion estimation time com pared with the UMHexagons, and cause little effect on the bitrate and picture quality.