Research On Reconstruction Algorithm Based On Structural Feature And Quantization Method In Compressed Video Sensing

Traditional video coding schemes are based on the Shannon-Nyquist sampling theorem.Specifically,the continuous video signal is first sampled at the sampling frequency twice as high as the highest frequency of the signal,and then complex coding techniques are used to remove the redundant information for lowering the bit rate.However,this coding scheme is of acquisition complexity and not suitable for resource-limited video capture devices,such as wireless video surveillance and wireless multimedia sensor networks,etc.Compressed Video Sensing(CVS)is a novel video signal acquisition scheme based on Compressed Sensing(CS)theory,in which the signal sampling and compressing procedures can be performed simultaneously.It is particularly suitable for resource-limited environments.Once raised,CVS has been widely concerned.CVS mainly includes four research aspects: measurement matrix,sparse representation,reconstruction algorithm and quantization method.In this paper,based on the analysis of the structure feature of video signal and the characteristic of CVS measurements,we deeply study the key technologies of reconstruction algorithms and quantization methods.The main work and research results are as follows:1.Based on the nonlocal similarity and the correlation among inter-frames in video sequences,this paper proposes an algorithm of structural similarity based inter-frame group sparse representation(SSIM-InterF-GSR),which effectively improves the reconstruction performance for compressed video sensing.In SSIM-InterF-GSR,the structural similarity(SSIM)is utilized as the block matching criterion to generate the group of similar blocks from the current frame and reference frames and the sparsity of the groups is used as the regularization term to reconstruct the current frame.Meanwhile,the step-decreasing scheme to determine the number of matching blocks is proposed during the iteration process of SSIM-InterF-GSR.The simulation results show that the proposed SSIM-InterF-GSR algorithm makes good use of the correlation among the video signal,prominently improving the quality of video signal reconstruction.2.The existing quantization methods for compressed video sensing measurements are mostly based on uniform scalar quantization with fixed bit-depth,ignoring the fact that the CS measurements or the measurement residuals(the difference between the current frame measurements and the measurements after quantization of the previous reference frame)are usually not uniformly distributed.Through the analysis of the measurement residuals distribution,a non-uniform scalar quantization scheme based on frame-based DPCM(DPCM-NSQ)is proposed for the CVS measurements.In addition,considering that quantization bit-depth has an important effect on rate-distortion(RD)in CVS,this paper proposes an adaptive optimal bit-depth estimation(AOBE)model based on DPCM-NSQ.In the model,quantization bit-depth is calculated by the function of predicted residual characteristic and sampling rate(SR).Experimental results demonstrate our proposed AOBE model has a superior RD performance compared with the fixed bit-depth method,while maintains almost the same complexity at sampling side.