块匹配电子稳像算法是一种稳定性好、准确度高的电子稳像算法.块匹配算法在目标区域中从起始点到匹配点进行搜索时，需要对图像块进行反复匹配，计算量大、实时性差成为限制其应用的主要问题.本文从缩小块匹配算法搜索范围的思想出发，提出了一种利用时间序列预测来确定最优搜索起始点的电子稳像算法.根据图像序列全局运动矢量的内部统计特性，选择合适的时间序列模型; 采用AIC准则和Durbin-Levinson递推算法估计模型的阶次和参量，并通过残差检验对模型进行检验和更新.利用建立的时间序列模型和历史数据对当前时刻全局运动矢量进行最优预测，并将其作为搜索起点来进行下一步精确搜索.实验结果证明，时间序列预测方法有效缩小了块匹配算法的搜索范围，使计算速度得到较大幅度的提高，并可直接推广到其它电子稳像算法中.

Block matching electronic image stabilization algorithm is a stable algorithm with high precision.Block matching algorithm reiteratively calculates the difference of the two image blocks,from the starting point to the matched point in the interesting area.Huge computering work is the main problem preventing the application of the block matching algorithm.Time series prediction method is used to estimate the optimal starting point,in order to reduce the search time.Proper time series model is selected according to the inner statistical properties of the global motion vector,then the model parameter is estimated and the model is checked by residual analysis.The time series model and historical data are used to predict the current global motion vector,which is set as the starting point.The experiments show that this method can reduce the hunting range effectively,making the matching process more quick,and it can be adopted by the other electronic image stabilization algorithms.