基于单演小波分析的微小运动放大技术研究

Research on Small Motion Magnification Based on Monogenic Wavelet Analysis

基于机器视觉的微小运动放大技术能将细小变化增强到肉眼可视的程度。传统放大算法依靠拉普拉斯金字塔分解获得图像相位信息,算法复杂耗时,大扰动时会产生大量噪声。论文结合Riesz变换和小波多尺度分析,提出一种微小运动放大算法,利用设计的单演小波快速分解视频帧得到相位和方向信息,提高视频处理速度。与现有放大算法进行实验对比,得到的客观评价指标表明该文算法能更准确地放大运动细节,同时大幅缩短视频处理时间。

Small motion magnification technology based on machine vision can magnify small changes to the degree that peo⁃ple can see.Traditional magnification algorithms rely on Laplace pyramid decomposition to obtain motion phase,which are complex,time-consuming,and with a lot of artifacts under large-scale motions in the video.In this paper,a new motion amplification algo⁃rithm is proposed by combining Riesz transform with wavelet analysis.The image is decomposed to obtain phase and direction quick⁃ly and the motion is magnified through magnifying the phase differences between image frames.Compared with the existing motion magnification algorithms,the objective evaluation index obtained by the experiment shows that the algorithm proposed can more ac⁃curately enhance the motion details,while greatly reducing the processing time.