摘要
在分析人眼视觉稳像原理和特性的基础上,提出了一种基于特征块匹配的视频稳定算法。该算法采用均匀分布的方式确定特征块。根据视频图像抖动特性对运动矢量进行预测,确定运动搜索的起始点。采用大钻石与小钻石搜索模板结合的方式,在阈值和搜索空间的约束下完成特征块匹配。然后通过运动平滑选择有效的运动矢量。利用仿射变换矩阵对原始图像进行配准,得出稳定视频序列。实验表明,该算法有效地稳定了图像序列,并且满足快速性和实时性的要求。
A video stabilization algorithm was proposed based on the feature block matching after analyzing the image stabilization mechanism and characteristic of human eyes.Uniform distribution mode was used to define the feature block.The motion vector was predicted according to the video jitter characteristic to decide the start point of searching.With the searching templates of both Large Diamond and Small Diamond,the feature blocks were matched under the restraints of threshold value and searching space.The effective motion vectors were chosen to achieve the motion smoothing.The original image was matched by the affine transformation matrix to obtain the stablized video sequence.The experiment results show that the proposed algorithm stablizes effectively the image sequence with required rapidity and real-time character.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2010年第2期529-533,共5页
Journal of Jilin University:Engineering and Technology Edition
基金
教育部新世纪优秀人才计划项目(NCET-04-0545)
国家自然科学基金项目(50875240)
关键词
工程仿生学
电子稳像
运动搜索
视频抖动
engineering bionics
electronic image stabilization
motion estimation
video jitter
