Research On Improvement And Application Of Robust Principal Component Analysis Model

With the continuous development of information processing technology,data acquisition technology,the arrival of big data era,people have more and more ways to obtain information,the amount of which is also increasing.As an important information carrier,videos and images are playing an important role in the field of image processing increasingly.Robust principal component analysis(RPCA)model is one of the most effective methods for image processing.This paper will mainly study the improvement of RPCA model and its applications in foreground-background separation.The main work of this paper is as follows:Firstly,considering the problems of the original RPCA model and most improved models that use nuclear norm to approximate rank function,resulting in too much computation and unsatisfactory foreground-background separation effect under dynamic background,a RPCA model with low calculation cost based on matrix factorization is proposed.By introducing matrix factorization technology,the problem of high computational cost caused by singular value factorization of large-scale matrix in the process of solving nuclear norm can be effectively solved.For full use of the time-spatial information of the original data,the binary mask is introduced to estimate the dense motion of the foreground with significant motion,and the confidence map is calculated through the standardized formula to obtain the rough position information of the moving object,respectively.The augmented Lagrange multiplier method is used to solve the improved new model,and experiments are carried out on a large number of real data sets.Experimental results show that the improved new model is effective for foreground-background separation.Secondly,considering the problems that the solution obtained by nuclear norm approximate rank function is only the sub-optimal solution of the original problem and the redundancy in large-scale data due to the slow movement of foreground objects,a non-convex RPCA model based on data preprocessing is proposed.The data with large scale,complex scene and high redundancy are preprocessing to eliminate some redundant frames information.Then,the rank function is approximated by introducing non-convex ?-norm to improve the approximation effect.The proposed new model is solved by augmented Lagrangian multipliers method and which effectiveness for foreground-background separation is verified through a large number of experiments,especially when the foreground object in the videos moves slowly.Thirdly,we summarize the work done in this paper and put forward further research directions.