Impulse Noise And Gaussian Noise Denoising Method

Image noise can be understood as an impediment to human visual perception, or a variety of factors that hinder the system image sensor to accept the source of information,or can also be understood as the deviation between the ideal signal and the true signal.A noise is generally a random signal which is unpredictable.We commonly use statistical method for its analysis.Noise of the image affects the entire process of the image processing that include the input, collection, processing and output of the results.In particular, the suppression of noise in image is a very crucial question.If the input accompanied by a large noise, the noise inevitably affects the entire process and output results.From the view of classification noise is varied. In summary, a noise is randomly generated, so it can use a statistical point of view mathematically to define the noise. Various types of noises are reflected in the image screen, and can be divided into two types:salt and pepper noise and Gaussian noise.This paper first describes the deterioration of the image model, afterward salt and pepper noise and Gaussian noise were introduced. Then a detailed description of the two types of image de-noising methods is introduced:spatial filter denoising and frequency-domain filter de-noising.At the same time commonly used evaluation methods of de-noising effect are introduced.This paper mainly consists of three parts.In the first section, we introduce some common Salt and pepper noise removal methods.Then we combine the merits of mean filters and adaptive median filters, and propose a new algorithm for the improved adaptive median filters.The experimental results show that the algorithm can eliminate high-density impulse noise in the image and preserve the details and edge information of the original image effectively.In the second part, we introduce the independent component analysis, preparing for the sparse coding in the third part. In relative gradient algorithms of independent component analysis, careful selection of step size is important to obtain good performance.In this section, an adjustable rate of relative gradient algorithm is proposed.With the changes of iteration number, the learning rate of relative gradient algorithm corresponding changes,which solve the problem on the contradiction between the convergence rate and stability well.On this basis, using this method in blind signal separation problems, simulations verified its useful behaviors.In the third part, we first introduce the Wiener filtering to remove Gaussian noise from contaminated image.Then using independent component analysis method for the Gaussian noise denoising, we propose a method of sparse coding image denoising that is based on polynomial fitting.In sparse coding, data vectors are expressed by a set of basis vectors.For each data vector, only a small portion of the basis vectors are activated at the same time.Shrinking the coefficients obtained from image data after sparse coding, shows better denoising effect. In this paper, polynomial fitting method is used to obtain sparse code shrinkage function. The method has good results.