Research On Signal Enhancement Based On Small Changes Of Video Signal And Its Application

Video is a common way for humans to record and perceive the world.We can extract many aspects of information such as voice and images from it.At the same time,video is dynamically changing,and the changes can be reflected by a series of image sequences on the timeline.In some situations,tiny changes have great value.However,due to the limitations of human eye's,these changes are ignored so that it is difficult to use them reasonably.The traditional signal enhancement algorithm will amplify all the changes,but in more cases,we only hope to achieve signal enhancement for the changes of interest.The signal enhancement method studied in this thesis provides a convenient way to extract and magnify tiny motions from the video.And we designed an application scheme to verify the usability of the method,thus increasing the ability of the value of minor changes to be discovered and utilized.The focus of this thesis is how to extract the signals that have changed slightly from the video,and enhance the changed signals.In the process of the research,we use the literature method firstly to systematically collect and classify the literature,which related to the research content at home and abroad,and we gradually understand the latest progress of the research.Secondly,we use the comparison method to summarize a mainstream solution from multiple methods.Then the theoretical analysis and experiments are carried out on each step in the process of the motion magnification method,especially in the aspects of signal extraction.Finally,through the design of the scheme,the application value of the method of video micro-change signal enhancement is demonstrated.The main work of this thesis includes the following:(1)Propose a method to determine the region of small changes.In order to avoid background interference and reduce the consumption of resources in processing,we proposed two method:foreground mask and the method based on the histogram of direction gradient.These two methods can narrow the processing range and reduce the impact of non-focused motion on the processing result.(2)Propose a method to separate the small motion and large motion.The video usually contains large motion and small motion of interest at the same time.Using the difference between the two changes in signal amplitude,the large motion can be removed through a filter.Then we only enhance the small changes of interest,which can reduce motion artifacts in the output video.(3)Periodic signal processing and frequency calculation.In some applications,the signal changes have a certain regularity,but its period is difficult to judge directly.We use the periodic change model to process and analyze the signal.Then the frequency value of the small change signal is estimated by the power spectrum estimation method and the peak detection algorithm.(4)Video-based sleep staging scheme.Based on the results of previous basic theory and algorithm research,breathing signals and eye movement signals were extracted from the video,and we designed a scheme for determining the sleep stage based on the characteristics of these two signal changes.The results obtained through experiments show that the scheme has certain application value.