| With the rapid development of the remote sensing satellite industry,the earth observation video captured by remote sensing satellites has shown huge application potential in more and more areas.In recent years,the launch number of remote sensing satellites has increased year by year,and the proportion of video satellites has gradually increased.In order to get a higher revisit ability and a higher resolution of the ground,many satellite companies have successively launched a star chain plan composed of loworbit video satellites.Therefore,a near real-time monitoring ability of the same area can be achieved.However,constrained by distance and orbit,the data transmission between satellites and grounds has the characteristics of high cost,small bandwidth,and low reliability.With the development of video satellite,large amounts of video data and limited bandwidth resources will inevitably bring great challenges to the compression capability of video data.Therefore,it has great significance to study video compression technology of earth observation video.The earth observation video data captured by remote sensing satellites is different from ordinary video in the ground.Constrained by the orbit,it has characteristics of overlooking from high altitude,large background,slowly changed landscape,multiple satellites revisit and so on.These features lead to a large amount of background redundant information among the revisited satellite videos.Repeated encoding of this redundant information between different videos causes a great waste of space bandwidth resources.Thus,unlike the classic video coding framework,which mainly focuses on local spatio-temporal redundancy during short-term,this study takes long-term background redundancy among revisited satellite videos during a long-time scope as research object.Firstly,aiming at the background similarity and modal difference between revisited videos,the mechanism of global redundancy of background data is studied,and a modal unified long-term background knowledge expression method is constructed;Secondly,in view of the problem of global motion and local color difference between the satellite video and long-term background,the inter prediction method based on global motion compensation and local color modal compensation is studied;Thirdly,for the problem of large differences in color modal of complex texture regions in the long-term background reference(LTBR),an intra prediction method constrained by prior knowledge of texture structures is studied;Finally,for the fusion problem of the long-term background redundancy removal method and the traditional short-term spatio-temporal redundancy removal method in this study,the fusion coding framework is studied to remove redundant information from multiple dimensions to improve coding efficiency.The innovations and achievements in this paper are as follows:(1)Long-term background knowledge redundancy and long-term background reference construction methodAiming at the problem of background similarity between revisited satellite videos at long intervals,which results in a large amount of background redundancy between massive revisited data,the mechanism of global redundancy of background data is studied,and a new idea for background redundancy removal based on long-term background knowledge dictionary is proposed.Also,the mechanism of the formation of modal differences at long-term intervals is studied,and a construction method of long-term background reference for the current encoded video is proposed.Experimental results show that the long-term background reference model constructed in this paper has a high degree of similarity and modal consistency with the current encoded frame.(2)Improved inter prediction method based on long-term background referenceAiming at the problem of global motion and local color difference between longterm background and the current encoded video,the compensation method of global motion and local difference is studied.Then an improved inter prediction method based on global motion compensation and local color modal compensation is proposed.The method improves the redundancy removal ability between the encoded frame and the long-term background reference.Experimental results show that this method has an average decrease of-14.74 % on the BD-Rate index compared to HEVC,and an average increase of 0.51 d B on the BD-PSNR index.(3)Intra-frame prediction method based on long-term background local structure knowledge constraintsAiming at the problem that regions with complex texture in the LTBR have large differences with encoded frame in color modal,the redundant expression method for structural knowledge in the long-term background is studied,and an intra-prediction method based on prior structure knowledge constraints is proposed.The method can effectively remove redundancy from LTBR under the interference of the local color modal differences,and further improves the efficiency of video coding based on LTBR.The experimental results show that this method has an average decrease of-23.10% on the BD-Rate index compared with HEVC,and an average increase of 0.72 d B on the BD-PSNR index.(4)Fusion coding framework to remove long-term redundancy and short-time spatio-temporal redundancyAiming at the fusion problem of the long-term background redundancy removal method and the traditional short-term spatio-temporal redundancy removal method,the best prediction mode selection method based on rate-distortion optimization is studied,and a fusion coding framework is proposed.The fusion coding framework can remove redundancy from both long-term and short term dimension,and highly improve coding efficiency.The experimental results show that the fusion coding framework has an average decrease of-31.69% on the BD-Rate index compared to HEVC,and an average increase of 1.11 d B on the BD-PSNR index.In summary,under a higher compression demand of satellite video for real-time data transmission from space to ground,this work studies long-term background redundancy in revisited satellite video,a satellite video coding method base on long-term background reference is proposed.Research results in basic theory and key technologies can provide new research ideas for satellite video encoding. |
PDF Full Text Request