Font Size: a A A

Video Deblurring Based On Deep Learning

Posted on:2020-08-13Degree:MasterType:Thesis
Country:ChinaCandidate:L Y WuFull Text:PDF
GTID:2428330590983163Subject:Control Engineering
Abstract/Summary:PDF Full Text Request
With the popularity of smart shooting devices and the rapid development of Internet technology,video sharing has gradually become an important way for people to communicate daily.However,in the process of shooting,due to external factors,such as cameras shake,high-speed movement of objects,etc.,the captured videos are blurred,resulting in a decline in video quality,which is not conducive to user viewing and information exchange.Therefore,video deblurring has become a key issue in the field of image processing,and it has very important practical significance.In recent years,with the great success of deep learning in the field of image processing,deep learning video deblurring has achieved better results than traditional video deblurring.Compared with traditional video deblurring algorithms,deep learning no longer requires manual design features,the network can be abstracted by extracting original image features.Therefore,around the problem of video deblurring,this paper starts from the perspective of deep learning,and builds the overall algorithm framework through deep neural network to achieve the purpose of video deblurring.Firstly,in view of the fact that the current deep learning video deblurring network structure is simple and the feature information is not fully utilized,the paper proposes a video deblurring algorithm based on DenseNet network.The algorithm DenseNet is the basic network,and the overall algorithm framework is built according to the encoderdecoder network style.At the same time,in order to make better use of the upper and lower frame information,the algorithm combines the partial training features of the previous frame and the features of the next frame training.The experimental results show that the video deblurring algorithm based on DenseNet network has better recovery effect than other mainstream video deblurring algorithms.Secondly,in order to achieve a more detailed deblurring effect,the paper proposes a video deblurring algorithm based on multi-scale recurrent network.The algorithm makes use of the existing multi-scale network from coarse to fine,and adds recurrent connections between networks of different scales,so that the weights are shared among different scale networks,and the learning ability of the model is more stable.Experiments show that the multi-scale recurrent network proposed by this algorithm can effectively improve the deblurring effect.Finally,through the results of the above two algorithms,this paper proposes a multiscale recurrent network video deblurring algorithm based on DenseNet.The algorithm combines the advantages of DenseNet and multi-scale recurrent networks.The experimental results show that the deblurring results of this method have achieved the best results in various evaluation indicators,which fully proves the robustness and effectiveness of the model.The research in this paper is aimed at practical problems,which not only enhances people's experience,but also provides effective protection for subsequent processing and transmission of video information.
Keywords/Search Tags:Deep learning, Video deblurring, DenseNet, Multi-scale recurrent network
PDF Full Text Request
Related items