基于最优运动矢量预测过程的改进与优化

Improvement and optimization of optimal motion vector prediction process

为解决UmhexagonS算法在预测最优运动矢量过程中运动估计时间消耗较大的问题,提出一种改进的情景分类(situation classification UMHS,SCUMH)算法。在起始点搜索环节得到最优点后,直接进入EDR模型的三步迭代运动情景分类判定;对大范围搜索模型进行矢量预测,根据最优预测运动矢量落入范围采取1/8区域划分搜索;在5×5搜索模型中,根据运动矢量的分布特性采取一种由内向外扩展的搜索顺序,节省了运动估计时间。实验结果表明,在运动估计时间方面,SCUMH算法比UmhexagonS算法节省了33.48%,在微运动情景下节省了27.05%,PSNR与码率基本不变。

To solve the high time consumption of the UmhexagonS algorithm in the optimal motion vector estimation,an improved situation classification algorithm was proposed.After obtaining the best point in the searching period,three-step iterative motion scene classification of the EDR model was applied directly.The 1/8 area division searches method was used according to the best predictive motion vector dropping range.Basing on the distribution characteristics of the motion vectors,a searching sequence from inward to outward was taken,which saved the motion estimation time in the 5×5 search model.Experimental results show that the proposed solution can save 33.48% of motion estimation time and about 27.05% of that in the slow motion situation in comparison with the UmhexagonS algorithm,and the PSNR and the bit rate are essentially unchanged.