中文摘要：

针对Type-1视频编码平台中的帧间预测仅有全搜索算法,从而计算复杂度很高的问题,提出了改进的UMHexa-gonS快速搜索算法.该算法采用了UMHexagonS算法的框架,针对该算法在参考帧数目和编码模式受限的条件下,建立了一种起始位置运动矢量预测模型;同时,为了解决UMHexagonS算法中第2类提前截止阈值计算不准确而导致编码性能下降的问题,提出了利用空间相关性的阈值修正方式.在Type-1平台中的实验结果表明,该算法的搜索精度较高,并且能够较好地适应不同序列的纹理特性.相比于全搜索算法,平均节省97%以上的时间,同时编码效率下降控制在平均0.032 dB以内.得出结论：改进的算法能够提升编码效率,同时节省平均运动搜索时间.

英文摘要：

In order to reduce the search complexity,this paper presented a novel motion estimation algorithm based on the original UMHexagonS.Compared with the original algorithm,the proposed UMHexagonS utilized a novel starting point prediction model to compensate the constraints of limited reference pictures and prediction modes.In the meantime,the proposed method modified the type-2 threshold used in the original UMHexgonS,which would cause great efficiency loss due to its non-adaptive calculation.The modification involved the threshold correction based on spatial correlation of the neighbor motion costs.Experimental results show that the proposed method can accurately reach the best search point and adapt to textural features of different sequences.Compared with the full search strategy,the proposed algorithm saves more than 97% of the search time while limiting the average coding efficiency loss to 0.032 dB;compared with the original UMHexagonS,the algorithm has advantages in both coding efficiency and average motion search time.