Research On Improved Reconstruction Algorithm Of Distribured Compressive Video Sensing

Along with the rapid development of multimedia technology,the video real-time transmission technology of large data volume has received extensive attention from scholars in domestic and foreign fields.As a multimedia communication technology,video communication is widely used in video scenes such as air and space transmission.Video signals has the characteristics of intuitive,accurate,and efficient.However,due to the large amount of video data,transmitting signals containing video information also requires a high channel bandwidth,which puts tremendous pressure on the transmission equipment.Distributed Compressive Video Sensing?DCVS?technology combines the advantages of Compressive Sensing?CS?theory and Distributed Video Coding?DVC?technology,which not only breaks through the Nyquist sampling theorem,but also the sampling requirements transfer the computationally complex and time-consuming motion estimation and compensation operations in the traditional video coding technology to the decoder,which greatly reduces the computational pressure of the coding end,and is very suitable for node distribution asymmetry video transmission scenario.In order to further improve the quality of system reconstruction and improve the overall performance of the system,this paper improved the key frame reconstruction algorithm,side information generation algorithm and non-key frame reconstruction algorithm in the system decoder.For the key frame reconstruction algorithm,the BM3D denoising model is introduced on the basis of the approximate message passing framework?AMP?.A new side information generation algorithm HEVC-ME is first proposed,compared with the traditional motion estimation.Algorithm,HEVC-ME algorithm selects block matching search with different size of coding units according to the characteristics of video frames.It added motion vector prediction technology in the process of search.The obtained side information frame more accurately approximates the original video frame.The reconstruction of non-key frames provides more accurate prior information.For the reconstruction of non-key frames of the system,the reconstruction model ofl₁-l ₁ minimization is proposed for the first time.This method is jointly reconstructed by combining the non-key frame compression results and residual value of the non-key frame compression result with the sampled side information frames.And by different compression ratios,for non-key frames,the relative accuracy of the compressed result and the sampled side information frame is given different weight coefficient during reconstruction.The simulation results show that compared with the previous DCVS reconstruction schemes,the proposed algorithm can improve the PSNR value of non-key frame reconstruction by 1 to 8 dB,and the overall system reconstruction gain is improved by 0.2 to 3.8 dB,which significantly improves the reconstruction performance of DCVS system.