Spatial Error Concealment Algorithm Research Based On Coupled Sparse Optimization

With the continuous development of digital media technologies,multimedia image videos have been applied in all aspects of work and life.High-definition video has greatly improved the user experience,but high-definition image video means higher data volume.Due to the limited current network bandwidth during transmission,the video stream is generally transmitted after being compressed.Compressed bitstreams mean worse error recovery capability when packets are lost,and the recovered images at the receiving end will be severely distorted.One way to solve this problem is to retransmit the video,but this method means high real-time performance and bandwidth cost.The error concealment algorithm can take advantage of the algorithm at the decoding end without changing the structure of the encoder.We can recover lost information and ensure the quality of the receiver's image by means of the method,so it is widely used.Error concealment technology is divided into time domain error concealment,spatial domain error concealment and spatio-temporal domain error concealment based on the redundancy information utilized.Because time domain information is difficult to obtain,this paper focuses on spatial domain error concealment technology.The core idea of the spatial domain hiding algorithm is based on the characteristics of neighborhood information correlation,using the surrounding information for linear interpolation to recover lost pixels.Existing spatial domain error concealment algorithms are classified into linear interpolation methods and sparse expression methods.However,when using the linear interpolation algorithm to recover the unsmooth images,there is a great inconsistency in the neighborhood information that will result in the blurred image.Therefore,this paper mainly relies on the sparse expression algorithm,and analyzes and improves the template matching module and local linear correlation analysis model in the dictionary construction phase.The first part of this paper optimizes the template matching module for the dictionary construction phase.Dictionary construction is the key to sparse representation and sparse solution.Template matching is to prepare the data needed for the dictionary.The traditional template matching algorithm uses a fixed threshold as a data selection criterion to construct a dictionary.Such a fixed threshold-finding matching block may result in inaccurate partial matching results due to lack of flexibility.Aiming at this problem,this paper proposes a template matching algorithm based on dynamic thresholds.According to the feature differences between pixel blocks to be restored,adaptively calculate the template matching thresholds,so as to select a more accurate observation set for the pixel block to be restored.And potential set,so get a better reconstruction effect.This paper tests the proposed method through three scenarios.For the independent loss case,the method improves the PSNR by an average of 0.29dB compared with the Baseline method.For the continuous loss case,the PSNR improves by an average of 0.28dB.For the random loss case,the PSNR has been improved by 0.07 dB on average.The second part of this paper is about the optimization adjustment of local linear correlation model.Sparse representation requires reconstructing information as a precondition when solving coefficients,and local linear correlation model is mainly used for rough construction of reconstructed information.However,the traditional local linear correlation model is based on the ridge regression method,and the ridge regression method is sensitive to the parameter setting when the reconstructed signal is roughly estimated.When the model parameters are small,the model seems to be more stable and stable,but the results of the model solution may be numerical stability and sacrifice a certain degree of unbiasedness.Aiming at this problem,this paper proposes a local linear correlation model based on canonical correlation analysis.The canonical correlation analysis algorithm can significantly reveal the internal relationship between two groups of variables,and the dimensions of the two groups of variables to be processed can be changed arbitrarily.,More flexibility.This paper tests the proposed method through experiments in three scenarios.The experimental results show that for the case of independent loss,the PSNR is improved by an average of 0.55dB in comparison with the method with the best airspace error concealment effect.Methods Compared with the best method for concealing spatial errors,the PSNR has been improved by 0.54 dB on average.For the random loss case,the PSNR has been improved by 0.80 dB on average.