Research On The Method Of Image Additive Noise Type Recognition And Parameter Estimation

It is difficult to avoid noise pollution in the process of generating,transmitting,compressing and receiving.Noise not only affects the quality of the image,but also limits the subsequent operations such as design of the corresponding filter,image segmentation,compression,restoration,and feature recognition.Therefore,the recognition and estimation of the image noise is not only the prerequisite for the research of noise suppression schemes,but also plays an important role in the subsequent image processing.The existing denoising methods are based on various research fields and methods are also varied.But each method has its advantages and disadvantages,and all of them have the appropriate scope of application.Some denoising methods are based on the pre-estimation of the size of the noise,such as the denoising of some wavelet thresholds.The research shows that the current popular denoising algorithm can reduce the noise performance greatly when the estimation of noise parameters is not accurate.Therefore,it is very important to identify the types of noise and estimate the size of the parameters.In view of the above problems,this paper has done the following research.First,the common noise distribution model is analyzed and identified,including the category of increasing noise and the improvement of the recognition method.These noises are mathematically modeled based on statistical methods,and random matrixes of different types of noise are obtained,and these noise matrixes are loaded into the grayscale images.Then the histogram is sampled and drawn from the discontinuous regions with relatively consistent gray level.The noise types are identified through the distribution and shape of the gray values of the different components in the histogram.On the basis of predecessors,we added Rayleigh noise,gamma noise,exponential noise and other types of recognition,and improved the method of drawing the gray-scale histogram,which extended the optional range of the sampling area.At the same time,it can intercept the different uniform discontinuous regions of the same adding and adding noise,and the accuracy of the noise type recognition can be improved by multiple sampling.Second,the influence of the rich texture features of the natural image on the coefficient distribution of the wavelet domain and the estimation of the size of the noise parameters are analyzed.On this basis,this paper proves that the mode of local variance distribution of diagonal wavelet can be used to estimate the noise parameter.But the diagonal wavelet coefficients contain not only the noise coefficient,but also some information of the original image.Therefore,in order to get to the noise of wavelet coefficients is pure the slot belt,and information on the part of the original image with diagonal wavelet coefficients in the estimate.Finally,subtract the estimated coefficients of the original image with the diagonal wavelet coefficients of the noisy image to obtain a relatively pure noise figure.The estimate of the local variance is the estimate of the noise standard deviation.Experiments show that this method can obtain more accurate noise estimate.Especially in the case of less noise,more detailed image information,the effect is more obvious.At last,taking several typical denoising methods in the wavelet domain as examples,how the estimation of the type of image noise and the size of the parameters is applied to the denoising method is analyzed in this paper.Several common denoising algorithms that need threshold setting are analyzed.In these algorithms,the standard deviation of noise is estimated in advance.