预决策金字塔层数的欧拉视频微弱运动放大算法  

Pre-decision Pyramid Layers of Eulerian Video Magnification Algorithm for Subtle Movements

作　　者：王霞[1,2] 付晓静[1] 王楠[1] 王蒙军[1] 

机构地区：[1]河北工业大学电子信息工程学院 [2]天津市电子材料与器件重点实验室,天津300401

出　　处：《科学技术与工程》2016年第1期114-118,共5页Science Technology and Engineering

基　　金：河北省高等学校自然科学研究重点基金项目(ZD20131043);天津市自然科学基金重点项目(14JCZDJC32600)资助

摘　　要：欧拉视频放大算法(EVM)将受人眼局限而观察不到的微弱运动放大;但放大视频质量随使用的金字塔层数及微弱运动幅值的不同而有明显差异。为提高视频放大效果,放大之前对输入视频提取运动前景,找到黑色像素数最大图像帧,求出该帧非0像素数占图像帧总像素数的比例;根据阈值判断金字塔层数,再进行EVM处理。研究表明,根据视频运动前景中非零像素与总像素比例,预判断欧拉视频放大金字塔层数,有效改善了单一金字塔层数EVM算法的放大效果。Eulerian Video Magnification algorithm is used to amplify the subtle movements which are not easily observed due to the limitations of the human visual system. But the magnified video＇s quality are quite different not only along with the layers of the pyramid but also the amplitude of the small movements in video. Detecting the movements foreground for the video before the magnification,find the frame which have the maximum value of black pixel,then calculate the ratio t,which is the percentage of the number of non-zero pixels in the video frame to the total pixels. At last determine the suitable pyramid layers according to the threshold,then carried out the EVM process. The results show that the amplification effect of EVM with single pyramid layers are improved effectively,based on the pre-decision suitable pyramid layers in Eulerian Video Magnification according to the proportion of non-zero pixels to the total pixels in the foreground extraction.

关 键 词：欧拉视频放大 微弱运动 图像金字塔 前景提取 

分 类 号：TP394.41[自动化与计算机技术—计算机应用技术]