Plug-and-Play video reconstruction using sparse 3D transform-domain block matching

In this paper, we propose a novel video reconstruction methodology built based on a generalization of alternating direction method of multipliers (ADMM) named Plug-and-Play. The motivation of the proposed technique is the improvement in visual quality performance of the video frames and decreasing the reconstruction error in comparison with the former video reconstruction methods. The proposed algorithm is an end-to-end embedding tool to integrate video reconstruction techniques with denoiser methods. Correspondingly, we use compressive sensing (CS)-based Gaussian mixture models (GMM) as a sub-problem regarding the proposed framework which is used as a method to model spatiotemporal video patches for video reconstruction. On the other hand, sparse 3D transform-domain block matching is applied as the denoiser of the proposed methodology to remove the remaining artifacts and noise in the reconstructed video frames. Consequently, by considering both online and offline CS-based GMM frameworks, we are able to make two forms of GMM-based embedding video reconstruction algorithms. The outcome has been compared with the result of CS-based GMM algorithm, GAP, TwIST and KSVD-OMP on the same datasets considering PSNR, SSIM, VSNR, WSNR, NQM, UQI, VIF and IFC as the evaluation metrics. It has been experimentally proved that the proposed online and offline GMM-based Plug-and-Play algorithms have more suitable results in comparison with their conventional CS-based online and offline GMM counterparts as well as other state-of-the-art techniques. The general quantitative results (considering all the datasets) for online proposed method regarding PSNR and SSIM metrics are 29.84 and 0.891, respectively, which is higher than the results of other techniques.