Image Denoising Study Based On Low-rank Matrix Factorization With Prior Knowledge

With the development of science and technology and the advent of the intelligent era,pictures have become one of the important channels for people to obtain information.However,under the influence of transmission,compression or the external environment,the picture will be contaminated,so that the image data has noise,which directly affects the subsequent image processing task.Traditional low-rank matrix factorization is generally suitable for Gaussian noise or Laplace noise,sensitive to outliers or non-Gaussian noise.Through the research of image denoising methods,in this paper,two improved image denoising models are proposed to address the shortcomings of the low-rank matrix factorization image denoising methods based on prior knowledge,the main work is summarized as follow:(1)The classical low-rank matrix factorization model always assumes that the noise conforms to the Gaussian distribution or the Laplace distribution,and the fitting effect is significantly reduced in the face of complex noise in the actual problem.In order to better fit the complex noise and enhance the robustness of the low-rank matrix factorization model,we propose a low-rank matrix factorization with double Gaussian prior,we use the Gaussian mixture model to fit the noise to increase the range of the noise distribution fitted by the model,and introduce the double Gaussian prior into the traditional Gaussian mixture model to improve the stability of the model.Experimental results verify that the proposed model can effectively process data with complex noise and achieve better and robust denoising effect.(2)In practice,not all the noise contained in the image is strictly symmetrical,and the use of symmetric distributions usually results in a reduced fitting effect of the model to noise.In this paper,a low-rank matrix factorization with mixture of asymmetric Gaussian model is proposed.We effectively model the noise by using mixture of asymmetric Gaussian model,and at the same time,the matrix obtained by decomposition is given a Gaussian prior,and the expectation maximization algorithm is used to solve the parameters of the model.The experimental results verify that the proposed model has a good denoising effect for the data containing different types of noise in the actual problem.