Research On Low Quality Video Recovery Methods

With the rapid development of Internet and the popularization of shooting devices,video sharing is becoming more and more necessary for people's daily life.However,due to external factors,such as camera shake,noise cause the decline in video quality,which is not conducive to viewing and access to information.Therefore,there is technology for low quality video.In order to solve the problems of video jitter and motion blur,we focus on video image stabilization and motion blur removal technologies.In this thesis,we study video jitter with moving scene changes,video frame motion blur and local moving object blur in video frames.The main research results are as follows:1.For the jittered video with scene change,this thesis presents a video stabilization method using panoramic image.Different from the traditional image stabilization method,the proposed algorithm uses the original video sequence to generate wide-angle panoramic view,and through the video frame extraction,thus synthesizing the output stable video.Then in order to eliminate the non-smooth motion of the original trajectory caused by video jittering,a motion mode discrimination method based on motion vector statistics is proposed,and we choose the appropriate smoothing on this basis.The proposed method can effectively remove large jitter and avoid the post-processing such as cropping the video width.2.To solve the blurry problems in video frames,we present an improved block synthesis video deblurring method based on spatio-temporal consistency.The accuracy of the blur function estimation model is improved by strengthening the inter-frame motion correlation.To improve the efficiency of our proposal,an adaptive selection method of reference frame based on frequency domain segmentation is proposed.The proposed method can effectively remove the blur of the video and achieve better result and efficiency than the original block sythesis method based on the blur kernel.3.For the blurry moving targets in the video frame,this thesis presents a local moving target deblurring method.The method of measuring the blur degree of local area is established based on estimates of relative motion between frames.And then by extracting blurry moving targets,the object-based local motion deblurring method is proposed.Experiments show that the proposed method both gets a deblurring result and improves the efficiency.