SIFT特征匹配和VFC算法的电子稳像技术  

Electronic Image Stabilization Based on SIFT Matching and VFC

作　　者：姜哲颖 周华兵[1] 刘姣[1] 

机构地区：[1]武汉工程大学计算机科学与工程学院,湖北武汉430205

出　　处：《武汉工程大学学报》2015年第9期45-49,共5页Journal of Wuhan Institute of Technology

基　　金：湖北省教育厅科学技术研究项目(Q20151503)

摘　　要：为了处理较为模糊的抖动视频,提出了一种基于尺度不变特征转换(SIFT)的特征匹配和向量场一致(VFC)优化算法的电子稳像技术.该技术着重于研究视频的运动估计阶段,利用SIFT提取高独特性的特征点,根据特征点进行帧间匹配,结合VFC,通过向量场的学习将外点从内点区分开来,得到两帧图像中特征序列的平移轨迹,然后估算出运动补偿向量并校正每一帧图像的相对位置,从而输出稳定视频.实验表明,该电子稳像技术处理抖动视频时精度高和耗时短.To process the blurry shaky video, electronic image stabilization based on scale invariant feature transform（SIFT） matching and vector field consensus（VFC） was proposed, which is mainly applied to video motion estimation. First, SIFT was used for extracting highly distinctive invariant feature and setting up the initial feature matching. Then, by the vector field consensus algorithm, we distinguished inliers from outliers and obtained the characteristics of translational trajectory in sequence of two video frames. Finally, we estimated the motion compensation to correct the relative position of every video frame by trajectory. Experiment results show the advantages of the method in precision and efficiency.

关 键 词：运动估计 向量场一致性 电子稳像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]