中文摘要：

传统的块匹配视频稳像算法不能完全去掉运动前景影响，在处理过程中需要预先设定判定相似度的阂值，容易造成计算误差，搜索算法搜索位置太多，加大了运算量。新算法首先对相邻两帧图像的匹配块进行x和y方向独立的块搜索，自适应地确定最优匹配位置；然后根据每次搜索最优匹配位置剔除运动前景目标块，调整步长进行下一步搜索，直到搜索步长为零时停止；最后计算出相邻帧间的移动向量进行运动滤波，补偿抖动。处理结果显示，该算法可以在视频大幅抖动、含有运动目标时和全局运动时进行准确稳像。

英文摘要：

The traditional block matching video stabilization algorithms cannot remove the influence of mo- ving objects completely and need a preset threshold value to decide similarity in processing procedure, which generates calculation errors easily and has lots of searching positions, thus increasing the computation load. First, the proposed algorithm searches the matching blocks at adiacent frames in a： and y independent directions and determines the optimal location adaptively. Second, according to the optimal location the proposed algo rithm removes the moving object areas and adiusts step length to next searching until the searching step length is zero. Finally, the proposed algorithm calculates the motion vector of adjacent frames with motion filtering and compensates the dithering of video frames. The processing result shows that this proposed algorithm can stabi- lize the video with large size dithering, global motion and moving objects.