Study On Filtering Method Of Mixed Noise In Digital Image

As the visual basic on which human perceives world, image becomes an important method in collecting, expressing and conveying information. In various kinds of image systems, it is hard to avoid degrading image quality due to transmission and conversion processing of imaging collecting, copying, scaning, conveying and displaying. The degradation typically appears as image with blur, distortion and noise. So, it has very important meaning to suppress noise, improve image quality and restore image in order to get digital image with high quality. In the field of digital image processing, image denoising is one of the most important and basic research topic at all times.In the field of digital image processing, various kinds traditional denoising methods have been propsed and even been applied for years. Upon such science background, the significance of making further study on the digital image denoising lies in several aspects as follows. Firstly, in image denoising area, traditional methods are applied wildly but they are not beauideal, which appears that the denoising operation can cause losing of image details. Therefore, studying on new denoising methods or improving existing algorithms is of great significance. Secondly, different algorithms have different basis of mathematical theory and denoising effect. So searching for the inner principle and seeking the relevant relationship to get better denoising effect are important tasks. Thirdly, the research of image denoising encourages the promotion of other digital image processing taches.When evaluating denoising effect in this paper, the following factors are considered. Firstly, the denoised image should be smooth enough with no or less noise trace. Secondly, the denoised image should not be blurred due to loss ofdetails excessively. Thirdly, there exists no artificial noise caused by concrete denoising method. Fourthly, the noise added in the experiment should approach to the natural noise. Fifthly, the result value of PSNR (power signal to noise ratio) should be as high as possible.The main task is to study how to filter the mixed noise composed of gauss noise and impulse noise, for the two kinds of noise are typical noise models. The main study work on filtering the mixed noise of image includes three items, linear filter (such as FIR filter), nonlinear filter (such as median filter) and the main study method of hybrid filter.As a typical linear filter, FIR filter is fit to be applied to digital image denoising for the exact linear phase. In the article abundant experiment and analysis have been done on the arithmetic of the FIR filter. In these FIR filter simulation experiments, the capability of filtering the mixed noise has been tested and measured in comprehensive ways. Analyzed the experments and datum, conclusion can be drawn. The FIR filter is capable of filtering the mixed noise to a certain degree, improving the PSNR in some range. But the noise trace is remained in mass. FIR filter bears the merits and demerits of linear expression, including the merits of simple and mature mathematical expression, the simple design, easy realization and the algorithmic some ideal specificity. The demerits contain the disability of impulse noise's removal. Upon all the study about the FIR filter, a conclusion comes up that the FIR filter can filter the mixed noise in certain degree but the effect is not very ideal.Nonlinear filter is one important branch of nonlinear science. It can maintain the signal details in certain degree and simultaneously filters the noise. Distinguished to the linear filter, it does not have the superimposition characteristic and proportion characteristic. Therefore, the output signal is not the convolution of the impulse response function and the input signal. The medianfilter is the nonlinear filter also one kind of order-statistics filter. This article first studies it's filter effect on single gauss or pulse noise. Through the summary and the analysis of massive experimental images and the datum, conclusion is drawn as follows. The median filter has good effect on filtering mixed noise. The PSNR value enhances comparatively much. The filter effect on the impulse noise is a little better than on the gauss noise. And then the filter effect of hybrid filter is studied. The observation on the experiment image discovered that, the median filter can achieve the effective effect to different mixed noise. But the denoising image comes up with serious loss of detail information. Therefore generally speaking, the filter has some flaws but overall performs well.A new hybrid filtering method is proposed in this thesis, which combines the theory thought of FIR linear filter and median nonlinear filter. To illustrate the denoising performance of the hybrid filter, a large number of simulation experiments, which contains filtering of single noise and mixed noise and analyzes the experimental results, have been done. The experimental results show that the hybrid filter can suppress Gauss noise, impulse noise and mixed noise effectively, thus improve PSNR dramatically. Meanwhile, it can be seen from comparing a large number of experimental images that the proposed method can both enhance PSNR and protect edge information. Therefore, the conclusion can be obtained that the hybrid filter has a good filtering performance from the view of both subjective effect and object measure of image.The simulation results in this thesis show that all these three filters can in certain degree filter the mixed noise in digital image. Relating all the PSNRs of various methods in experiment to image effect to analyze, conclusions can be drawn as follows.The first, the denoised image filtered by the hybrid filter, in contrast to the original noised image, improves the PSNR greatly. And the mixed noise is filteredwith edge preferable protection. So the hybrid filter can be considered well in filtering performance.The second, the PSNR of the hybrid filter is higher than those of the other two kinds of filters, comparing the hybrid filter with the FIR filter and the median filter in filtering mixed noise. From contrasting subjective quality of experimental images, it can be learned that the hybrid filter is able to effectively remove image noise. Moreover, the detail protection of the hybrid filter is also the best of the three.The thesis contrasts the linear filter, nonlinear filter and the proposed hybrid filter by analysis and experiment of filtering mixed noise from digital image and validates the better filtering performance of the proposed hybrid filter for removing mixed noise in digital images. Also, the mathematical basis of the hybrid filtering method is explored and analyzed. These outcomes can provide theoretic and experimental foundation for further research.