Vision Enhancement And Its Application Based On Subtle Motion Magnification

Video motion magnification is a kind of techniques for changing the amplitudeofsignal of interestedmotionin videos. Using the method, we are able to reveal subtlechangesin the world that are originally invisible to naked eyes, so as to enhance the visualeffectand exploit valuableinformation.Depending on processing methods of motion, video motion magnification techniquescan be divided into feature-based Lagrangian motion magnification and pixel-basedEulerian video magnification (EVM). Among them, the Lagrangian methodamplifiesvariations by trackingfeatureof optical flow and increasing range of motion, andacquires synthetically magnified video with imageinpainting and rendering. But thismethodbrings computing tedious and practical difficulties.Eulerian method, by analyzingeach pixel gray value changes over time in video, usetemporal filter to amplify the signalof interested motion. But the method need to process three channels in the YIQ space,which is very time-consuming. Meanwhile, this method uses many experiments andexperience to select the frequency band of band pass filter, so that noise also enlarged inthe signal. Eulerian method use a fixed filter for processing the whole video sequence,which exhibit the same variation of amplified signal and cannot effectively retain thestate of motion of the original signal. And in the actual video processing, automatedfrequency detection and amplificationis particularly important.To solve these problems, this paper proposes a method based on Y channel toaccelerate and a small motion automated amplification method for a single frequency andmulti-frequency. And the automated amplification is achievedby frequency estimationmethod and frequency automated detection way to design an appropriate filter. Specifically,this paper has made the following contributions.1) This paper firstly discusses two motion amplification methodsthat based on RGBspaceand based on the Y channel in the YIQ space. The experimental resultsshow that:comparing with EVM method, the processing speed of RGB method has improved, but ithas much of a difference in the motion magnification effect; and comparing with EVMmethod, Y channel method got the obvious increase in processing speed, and theamplification effect is basic consistent.2) For a single frequencyvideo sequence,we analyzethe variation of motion signals inthe case of a low spatial resolution by a frequency estimated of global manner. And each frame of the video image is processed by Gaussian smoothing down-sampling.Thefrequency is obtained bytime-frequency analysisof frequency distribution of all thepixels.And the band-pass filter is designed for amplification process by cluster analysis,which can select the center frequency and distribution of the point of small motion.Experiments show that this method effectively suppresses noise signalamplificationbymagnifying the motion information of video, and it also provides apractical idea offrequency parameter selection problemin videos.3) For multi-frequency distribution over timein videos, we use the way of automateddetectionfrequency.According to S transform of each pixel, the instantaneous frequency atdifferent times can be obtained. Thus, the automated motion magnification is achieved bydesigning dynamic filter on the time series. Experimental results show that this method caneffectively deal with non-stationary video sequences.Because of the dynamic design of thefilter is not only to ensure that the anti-noise performance of the algorithm, but also tokeepeffectively the initial state of motion information changes.This paper presents an automated methodof subtlemotion magnification to processtheactual videos. It uses more convenient and effective operation for subtlemotionmagnification technology, which will lay amore widely used foundation.