Research On Distributed Video Coding Based On Video Features

Distributed Video Coding(DVC)is a new video coding method,which is suitable for some application scenarios where coding resources are more limited than decoding resources.Compressed Sensing(CS)is a new theory,which allows original sparse signals to be recovered from fewer measurements compared with traditional methods.Combined with sparse characteristics of video signals,Distributed Compressive Video Sensing(DCVS)codec system combines CS theory and DVC technology,the quality of signal reconstruction directly affects the performance of DCVS system.At present,there is still a large space for optimization of reconstruction algorithms.At the same time,the generation quality of side information has a great impact on the performance of DCVS system and the quality of decoded video frames.At present,there have been many achievements in the research on the generation algorithm of side information,but the generation quality of side information still has a great space for improvement.Based on the characteristics of the video signal itself and combined with CS and DVC technologies,this paper proposes a series of improvement schemes for the compressed sensing reconstruction algorithm and the side information generation algorithm,and the work is as follows:First,study the existing DCVS system solutions,in compressed perception part focuses on the measurement matrix,discusses and analyzes the structured random matrix,through the comparative experiments with other measurement matrix,matrix has been verified with gaussian unrelated features of the matrix,calculation efficiency,smaller storage space requirements,suitable for use in image and video signal compression sampling.Secondly,common compressed sensing reconstruction algorithms are analyzed,and the following deficiencies are found through simulation experiments: for example,greedy reconstruction algorithms require known signal sparsity,which cannot be known in advance in actual scenes;At the same time,the atom selection of signal support set needs to be optimized,and the convergence speed and accuracy of reconstruction algorithm need to be improved.Based on the shortcomings of the reconstruction algorithm mentioned above,an improved reconstruction algorithm is proposed: the idea of backtracking regularization is applied to the selection of support set,and the variable step size module is used to support sparsity adaptive to improve the performance of the algorithm.Then,the paper analyzes the common side information generation algorithm,the traditional algorithm,which make use of video time correlation between adjacent frames for motion prediction.When processing some videos with complex motion,the performance of the traditional algorithm needs to be improved.Based on the shortcomings of traditional algorithms,an optimization scheme is proposed: the motion estimation and motion compensation algorithms are optimized by using the chromaticity characteristics of video and spatial correlation between adjacent frames to improve the generation quality of side information.Finally,combined with the improved compressed sensing reconstruction algorithm and side information generation algorithm,the improved DCVS system is simulated.Compared with the traditional scheme,the simulation results show that the reconstructed video frame quality is improved and the PSNR gain is about 1.64 d B?2.27 d B.