Research On Video Retargeting Based On Grid Deformation

With the diversification of digital display equipment,how to carry out the video resolution and aspect ratio to match the different display needs has become an urgent problem to be solved.The purpose of content-aware video retargeting algorithm is to guarantee the reconstruction quality of visual salient areas in video during scaling,while maintaining the temporal coherence of video frames.Existing video retargeting algorithms mainly use frame-by-frame or global methods to achieve video scaling.Frame-by-frame retargeting method has low computational complexity,but the retargeted videos are prone to more artifacts.And the retargeted video based on global retargeting is of good quality,but the computational complexity is very high.Therefore,the two methods need to be balanced effectively.In addition,fixing the size of each mesh is usually used by most of the existing video retargeting algorithms based on mesh deformation.Due to the diversity of video content,fixed mesh size usually affects the quality of retargeted video.In order to solve the above problems,the algorithm of video retargeting based on mesh deformation is deeply studied in this dissertation,which aims to improve its performance effectively.The results are described as follows.A video retargeting algorithm based on mesh deformation and frame grouping is proposed.The algorithm is based on mesh deformation and retargeting by frame grouping.The algorithm aims to effectively balance the advantages and disadvantages of frame-by-frame-based and global-based video retargeting methods.Specifically,in order to protect the aspect ratio of the visual salient region and the temporal coherence of the video as far as possible,the video frames are grouped according to the set mesh size on the basis of constructing the video camera motion model.Then the mesh is partitioned on the puzzle formed by each video group,and the scaling factor is obtained according to the corresponding temporal and spatial constraints of each mesh.Finally,the scaling factor corresponding to the mesh in each video group is used to retargeting all the video frames in the group,so that the video can be changed to any size.In this dissertation,the common test video sequences are used to test the algorithm.The experimental results show that,compared with the traditional video scaling method and other mesh-based video retargeting methods,the proposed algorithm has better subjective effect of retargeted videos.The temporal coherence of the original video is also effectively maintained.In order to solve the problem that the fixed size mesh affects the quality of retargeted video,a video retargeting algornthm based on objective quality evaluation and adaptive mesh partitioning is proposed.According to the intensity of video motion,the algorithm adaptively selects different objective evaluation indexes to evaluate the retargeted video with different mesh size,so as to output the retargeted video with better temporal coherence.Specifically,on the basis of constructing camera motion model,several different mesh sizes are divided,and different mesh sizes are used to group the video frames.Then,according to the number of groups,we select the objective evaluation of video saliency similarity measurement(SSM)or temporal inconsistency distortion(TID).Then the mesh is partitioned on the puzzle formed by each video group,and the spatial and temporal constraints on the mesh are carried out to obtain the retargeted video with different grid size.Finally,the retargeted video with different grid size is evaluated according to the selected objective evaluation index,and the retargeted video with better score is output.In this disseration,the common test video sequences are used to test the algorithm.The experimental results show that compared with the existing video retargeting algorithms,the proposed video retargeting algorithm has better subjective effect,while protecting the visual salience region of the retargeted video.It can maintain the temporal coherence of the original video more effectively.