Research And Application Of Distributed Video Compressed Reconstruction Technology Based On Spatio-temporal Correlation

With the rapid development of mobile Internet and multimedia technology,video applications are more and more widely used in mobile terminals,such as mobile videophone,wireless video monitoring,wireless visual conferencing and so on.In these video application scenarios,the mobile terminal has limited calculation,storage,and transmission capabilities,and it is difficult to handle complex operations.This makes the traditional video coding scheme with high sample rate and computational complexity at the coding side not suitable for such video applications.Therefore,in order to satisfy the new requirements under the new video application scenario,a distributed video compression sensing coding scheme is proposed,which not only has the characteristic of independent coding and joint decoding in distributed video coding,but also has the characteristics of low sampling rate,simple coding and strong anti-noise ability.it has great research significance and application value.At present,the research on distributed video compressibility mainly focuses on two aspects: 1)the first is how to weigh the computational complexity and coding efficiency of the coding side;2)the second is how to effectively use inter-frame and intra-frame correlation information to improve the efficiency and quality of reconstruction algorithms.This article mainly carries on the thorough research from above two aspects.The main work and the innovation content of the paper are summarized as follows:1.For the disadvantage of the traditional multi-hypothesis prediction reconstruction algorithm based on Tikhonov regularization that the prediction accuracy of video images is not high when the sampling rate is low at the encoder end.This paper proposes an iterative reweighted multi-hypothesis predictive reconstruction algorithm.The algorithm continuously optimizes multi-hypothesis set predictive coefficients by iteratively reweighting Tikhonov regular matrices,which improves the accuracy of multi-hypothesis set weighted reallocation and further improves the Reconstruct performance.The experimental results on the standard video sequence set show that the proposed algorithm outperforms the traditional multi-hypothesis prediction algorithm by 1~3dB(PSNR).2.Based on the above research work,in order to further improve the reconstruction,a distributed video compressed sensing reconstruction algorithm combining the best linear estimation and multiple hypothesis prediction is proposed.This algorithm achieves high-quality classification and reconstruction of non-key frames in video sequences through three mechanisms as similarity discrimination,measurement value supplementation and smooth discrimination.Experimental results on the standard video sequence set show that when the measurement rate is lower than 0.2,the proposed algorithm improves the PSNR of the average reconstructed video signal by 3 to 5 dB over a video sequence with many smooth or slow motions.3.A new distributed compressive sensing reconstruction algorithm based on residual-predictive reconstruction is implemented.The algorithm uses the sparse property of residuals of the predicted frame and the current frame,and achieves high-quality restoration of the video image data in the measurement domain by the gradient projection method.The algorithm fully exploits and utilizes the spatio-temporal correlation of video image signals and improves the quality of data reconstruction.The experimental results on the standard video sequence set show that the residual-prediction based reconstruction algorithm effectively improves the video image reconstruction quality at the decoder without increasing the coding complexity.