Research On Multi-hypothesis Prediction And Residual Reconstruction Algorithm For Compressed Video Sensing

Based on the Nyquist sampling theorem,traditional video coding program uses Nyquist sampling rate for high-speed video signal sampling.Through the complex motion estimation,motion compensation and sparse transform compression coding technology,it removes a large amount of redundant information of the video signal.In signal acquisition side,such coding scheme demands larger storage space and higher algorithm complexity.Therefore,it is not suitable for sampling resource-constrained application scenarios,such as wireless video monitoring,wireless multimedia sensor network,etc.Compressed Video Sensing(CVS)is a burgeoning video coding scheme in recent years.It realizes the signal sampling and signal compression simultaneously,greatly reducing the demand for collection resources.Therefore,CVS is especially suitable for the application scenarios with limited collection resources.In CVS,frames are observed independently and jointly reconstructed.Since optimal reconstruction performance is obtained in the reconstruction algorithm,the prediction-residual reconstruction model is studied in this article.How to adequately excavate the correlation information between video frames to improve the prediction precision of the current frame and how to get higher quality of residual reconstruction which has the character of low energy are two key points in the research of this paper.The main work and contribution of this paper are as follows:1.Through studying of the existing prediction and residual reconstruction algorithms,one can find that the processing of multi-hypothesis prediction and residual reconstruction based on initial reconstruct frame in pixel domain can effectively weaken the block effect,then improving the quality of reconstruction of video.The multi-hypothesis prediction residuals reconstruction iteration in pixel domain is proposed to improve the reconstruction quality of non-key frame in this paper.Considering strenuous movement scenes,some blocks may fail to find suitable hypotheses in reference frame or need a very large search window to get suitable hypotheses.In view of these two situations,the adaptive multi-hypothesis prediction with interframe or intra-frame model and region-joint fast search scheme are proposed in this paper.The simulation results indicate that the iterative prediction algorithm proposed in this paper is better to excavate the correlation information between video frames,and fast search scheme based on region-joint method improves the search efficiency.2.Smoothed Projected Landweber(SPL)algorithm is widely used in reconstruction of video residual with its characteristic of efficiency,but it introduces too much reconstruction noise when applied to residual reconstruction.Based on the SPL algorithm,the Overlapped Blocking SPL(OBSPL)residual reconstruction algorithm is proposed in this paper.Its sparse processing is proceeded in overlapping-blocks way.The OBSPL algorithm takes the advantage of local correlation between adjacent residual blocks.The average fusion processing for the overlapped part of residual blocks after sparse processing is helpful to eliminate the noise during reconstruction.The simulation results show that the proposed algorithm effectively restrains the reconstruction noise,and obtains good residual reconstruction performance.What's more,it has advantage of good portability.