依据图像序列的运动矢量的时空相关性和中心偏移特性,首先对宏块进行类型划分,并利用相关块而非相邻块进行初始点预测的选择预测方法,提出了一种对起点进行选择预测的自适应运动估计算法,而在搜索过程中,则根据不同的图像内容对不同的块灵活地采用不同的搜索模式。实验证明,该算法在大大提高搜索速度的同时,还能保证图像性能,还兼顾了运动估计复杂度和精确度的要求。

By studying the spatio-temporal correlation and the center-biased characeristic of motion vectors in image sequences,this paper classifies the macro-blocks as different types and takes advantage of the relative blocks instead of neighbour blocks to predict the initial point.A new adaptive motion estimation algorithm based on selecting predictive initial search point is presented,according to which different modes are agilely selected in accordance with image contents.It has been proved with the experiments that this algorithm greatly improves the searching speed and at the same time guarantees image performances with consideration to complexity and accuracy of motion estimation.