Speckle Noise Discrimination And Parameter Estimation Methods

Most digital images are often polluted by all kinds of noise when they are acquired and transmitted, which not only degrades image quality, but also hinders the consequent image analysis and image understanding, so to suppress noise is always one of the most important tasks in image processing. As a key step to noise reduction, a good algorithm should make full use of the prior knowledge, such as the type, intensity, composition of noise and so on, before an appropriate filter is designed. However, these prior knowledge usually come from some subjective assumptions, rather than the objective description of image noise, in this case, noise image processing is often blind. Therefore, it is urgent to estimate image noise in image processing.This paper combines Gray theory, parameter estimation, histogram information, Gaussian-Hermite Moments and curve fitting to analyze image noise. Then an identification method on noise type and a parameter estimation method on speckle noise are proposed. Based on these work, a demo software of image noise analysis and estimation is finally designed and developed. Main creative work can be summarized as follows:(1) Employing Gray relational analysis theory and histogram information, this thesis proposes an efficient method on noise type recognition, which is based on gray relational with information of histogram and can distinguish three familiar types of image noise, i.e., salt-and-pepper noise, Gaussian noise and Speckle noise. In the method, statistical information of noise-polluted image histogram is analyzed first. And then, salt-and-pepper noise is distinguished out directly. Second, Gray relational analysis theory is employed to analysis the histogram curve characteristic of Gaussian noise and Speckle noise. Finally, the two types of noise are identified from the different gray relational grades. Theoretical analysis and experimental results show that the proposed method is a simple and easy when compared with some existing methods.(2) An efficient method to evaluate the parameter of speckle noise is suggested, which is based on Gaussian-Hermite Moments. In the method, the characteristics of Gaussian-Hermite moments are discussed first, and then a feature vector based on Gaussian-Hermite moments is defined to analyze the distribution regularities of noise after speckle noise with different variances are respectively added to a certain subimage where the grayscale of all pixels is similar or the same. On the basis of the distribution regularities, noise characteristic value reflecting the intensity of image noise is determined, thus a function mapping of the noise parameter and noise characteristic value is deduced. Finally, an estimation function is established by curve fitting. Experimental results indicate that the estimation function can rapidly and correctly measure the intensity of speckle noise without any prior knowledge.(3) A simple and efficient method on blind parameter estimation of Speckle noise in images is presented. In the method, after various intensity speckle noise are added into an image respectively, the regularities of histogram distribution in unitary gray area are analyzed. On the basis of the regularities of distribution, noise characteristic value standing for noise intensity is constructed. And then, a function map between noise parameters and noise characteristic values are deduced out. Experimental results indicate that the function map can estimate the noise intensity in noise-polluted images rapidly and efficiently without any prior knowledge.Based on the above work, a demo software is designed and development via visual C++ 6.0, which not only can identify the image noise type, but also may provide with estimated noise parameters.