Research On Residual Signal Reconstruction In Compressed Video Sensing

The traditional video coding technology based on Nyquist sampling theorem first samples at Nyquist sampling rate to acquire video signal and then compresses the acquired video signal.This “sampling at high rate first and then highly compressing” style wastes too much sampling resources.In recent years,the field of distributed video processing has developed rapidly,where resource-constrained sampling facilities are adopted in the actual application circumstances,for example,wireless multimedia sensor network,and the traditional video coding technology does not fit any more.Compressed sensing theory,which can incorporates sampling with compression and has a simple encoder,has bright application prospect for resource-constrained circumstances,thereby has drawn great attention of academic community.In compressed video sensing(CVS)where compressed sensing theory is used to acquire video signal,the algorithms based on predictionresidual reconstruction framework gain the highest reconstruction performance,but the lack of in-depth study into residual reconstruction restricts the contribution of residual reconstruction to final reconstruction performance.In this paper,the deep analysis and study of the prediction residual is conducted,a block classifying reconstruction algorithm based on residual structure characteristics(BCSC)and a multihypothesis-based residual reconstruction scheme(MHRR)are proposed respectively.The detailed research work is divided into two parts as below:1)The algorithm widely used for residual reconstruction is BCS-SPL,which is originally proposed for natural images.Since prediction residuals differ from natural images in structure characteristics,the BCS-SPL algorithm cannot reconstruct the prediction residual very well.Through the in-depth research,this paper proposes a block classifying reconstruction algorithm based on residual structure characteristics(BCSC),which classifies residual blocks according to their average energy first and then adopts suitable reconstruction algorithms to each type of residual blocks.Simulation results show that,the proposed BCSC algorithm can achieve higher reconstruction performance than BCS-SPL algorithm for video sequences with fast movements.2)To further improve the sparsity of the prediction residual,this paper proposes a multihypothesis-based residual reconstruction scheme(MHRR).In the proposed MHRR,a method of generating hypothesis blocks set in residual domain is offered,and the pixel-domain motion estimation is novelly combined with calculating the linear prediction weights in measurementdomain to exploit correlation between residual frames efficiently.Simulation results show that,the proposed MHRR algorithm can achieve higher reconstruction performance than BCS-SPL algorithm for state-of-the-art CVS reconstruction schemes.