基于SPNet的视频运动放大方法  

Video motion magnification method based on SPNet

作　　者：周飞 杜振龙 Zhou Fei;Du Zhenlong(School of Computer Science&Technology,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学计算机科学与技术学院,南京211816

出　　处：《计算机应用研究》2020年第12期3820-3824,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61672279);江苏省“六大人才高峰”资助项目(2012-WLW-023);水文水资源与水利工程科学国家重点实验室开放基金资助项目(2016491411)。

摘　　要：视频运动放大技术在很多领域应用广泛,但是由于关注的运动过于微弱,很容易与视频噪声混淆,容易产生伪影和模糊噪声。为了克服传统算法在细节上的缺失和基于学习的算法在放大倍数上的限制,将手动设计滤波器与课程学习策略相结合,提出了基于可控金字塔分解的网络结构(SPNet),它在处理大幅度运动场景的同时也能具有更大的放大范围,且能保留更多细节。结果表明,在图像细节上优于传统算法,并且与基于机器学习的视频运动放大方法相比也有很大的优势。Video motion magnification technology is widely used in many fields,but because the motion of attention is too weak,it is easy to be confused with video noise,and it is easy to generate artifacts and fuzzy noise.In order to overcome the lack of detail in the traditional algorithm and the limitation of the learning-based algorithm on the magnification,this paper combined the manual design of the filter with the course learning strategy,and proposed a network structure based on the controllable pyramid decomposition,called SPNet(steerable pyramid networks).It could process a large motion range while also having a larger magnification range and could retain more details.The results of this method show that the image details are superior to the traditional algorithms and have great advantages compared with the machine learning based video motion magnification method.

关 键 词：运动放大 课程学习 可控金字塔 机器学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]