计算机应用 ›› 2017, Vol. 37 ›› Issue (5): 1460-1465.DOI: 10.11772/j.issn.1001-9081.2017.05.1460

• 计算机视觉与虚拟现实 • 上一篇    下一篇

基于S变换的非平稳微小运动自动放大

雷林1,2, 李乐鹏1, 杨敏1, 董方敏1, 孙水发1   

  1. 1. 湖北省水电工程智能视觉监测重点实验室(三峡大学), 湖北 宜昌 443002;
    2. 广东省粤电集团有限公司 天生桥一级水电开发有限责任公司水力发电厂, 贵州 兴义 562400
  • 收稿日期:2016-10-10 修回日期:2016-12-15 出版日期:2017-05-10 发布日期:2017-05-16
  • 通讯作者: 孙水发
  • 作者简介:雷林(1990-),男,贵州兴义人,硕士,主要研究方向:视频处理;李乐鹏(1990-),男,湖北汉川人,硕士,主要研究方向:视频处理;杨敏(1992-),男,湖北黄梅人,硕士研究生,主要研究方向:视频处理;董方敏(1965-),男,湖北荆门人,教授,博士,主要研究方向:智能信息处理;孙水发(1977-),男,江西黎川人,教授,博士,主要研究方向:图像处理、计算机视觉。
  • 基金资助:
    国家自然科学基金资助项目(61272237,61402259);湖北省自然科学基金创新群体项目(2015CFA025);湖北省教育厅科学技术研究计划重点项目(D20151204);水电工程智能视觉监测湖北省重点实验室开放基金资助项目(2014KLA04)。

Non-stationary subtle motion magnification based on S transform

LEI Lin1,2, LI Lepeng1, YANG Min1, DONG Fangmin1, SUN Shuifa1   

  1. 1. Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering(China Three Gorges University), Yichang Hubei 443002, China;
    2. Tianshengqiao-I Hydro-Power Development Limited Company Hydropower Plant, Guangdong Yudean Group Company Limited, Xingyi Guizhou 562400, China
  • Received:2016-10-10 Revised:2016-12-15 Online:2017-05-10 Published:2017-05-16
  • Supported by:
    This work is partially supported by the National Natural Science Foundation of China (61272237, 61402259), the Hubei Natural Science Fund for Innovative Research Groups (2015CFA025), the Major Program of Educational Commission of Hubei Province of China (D20151204), the Opening Fund of Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering (2014KLA04).

摘要: 现有的欧拉视频微小运动放大方法没有考虑视频中运动信息的自动检测,在实现运动放大时需要人为选取感兴趣运动信息处理的合适参数,比如滤波器截止频率、放大倍数等。对于一般的视频,通常无法直观地确定这些参数,而是通过试错来完成。为此,基于S变换提出了一种视频序列中非平稳微小运动的自动放大方法:基于S变换自动确定带通滤波器的即时相关参数并设计了相应的动态滤波器,在此基础上实现了视频微小运动的全自动放大。首先,通过S变换分析出视频中微小运动在各个时刻的瞬时频率;然后,对视频在不同时刻呈现不同频率的情况进行动态带通滤波处理;最后,放大经过带通滤波后的感兴趣运动信息,实现微小运动放大。此外,对于抗噪性能分析,提出了一种视频区域信噪比评价方法。实验结果表明,所提方法在对实际视频进行放大处理时,能够根据运动信息频率的变化自动地获取滤波器、放大倍数等参数,无需人工参与;运动放大后可以动态地展现运动目标的放大效果。同时,准确的动态滤波器在一定程度上能够抑制噪声干扰,使得运动放大效果更加理想。

关键词: 欧拉视频放大, S变换, 自动检测, 动态带通滤波, 信噪比

Abstract: The existing Eulerian video subtle motion magnification method does not take into account the automatic detection of motion information in the video, the appropriate parameters for motion information processing need to be selected in the realization of motion magnification, such as the filter cut-off frequency, and magnification. For the general video, these parameters usually could not be determined directly, but by trial-and-error. In this paper, an automatic amplification method of non-stationary subtle motion in video was proposed based on S transform. The instantaneous correlation parameters of band-pass filter were automatically determined based on S transform and the corresponding dynamic filter was designed. On this basis, an approach of automated subtle motion magnification was achieved. Firstly, the instantaneous frequency of subtle motion was obtained by S transform in video. Then the dynamic band-pass filter was used to process different frequencies at different times. Finally, the effective motion information which was filtered by band-pass filter was magnified to achieve subtle motion magnification. In addition, for the analysis of anti-noise performance, a method for evaluating the signal-to-noise ratio in video area was proposed. The experimental results show that the proposed method can automatically obtain the parameters such as filter and magnification according to the change of the frequency of the motion information when the actual video is amplified, without manual participation. After the motion amplification, the amplification effect of the moving target can be dynamically displayed. At the same time, the accurate dynamic filter can suppress noise to some extent, and makes the motion magnification effect better.

Key words: Eulerian Video Magnification (EVM), S transform, automated detection, dynamic band-pass filter, Signal-to-Noise Ratio (SNR)

中图分类号: