Speckle Reduction Algorithm Based On Two-sided Generalized Nakagami Distribution For Medical Ultrasound Images

As an important information source, image is easy to be influenced and interrupted by various kinds of noise in the process of acquisition, transmission and storage. The medical ultrasound (US) image has already become one of the most important tools for medical diagnosis because of its low-cost and portability. However, the inherent speckle noise in US image severely degrades image quality, and affects the following image processing as well, such as feature extraction, analysis and diagnosis. Therefore, one of the most crucial tasks in medical US image processing is to balance the preservation of image details and features against speckle suppression.Speckle reduction algorithm based on wavelet transform for medical US image has its significant meaning. In this thesis, the main work is introduced in the following aspects:1. Research of wavelet thresholding method is a classical one for image denoising. At first, compared with Fourier transform, the merits of wavelet transform are analysed, then the development and current situation about wavelet denoising are described in this thesis. In addition, the theoretical knowledge of wavelet denoising is briefly introduced, which establishes a great theoretical basis in wavelet domain image denoising. Thanks to decorrelating property of coefficients after wavelet transform, most of image details and features are preserved in a minority of coefficients with large amplitude. The wavelet thresholding method is pretty simple, while it has great denoising performance. For this reason, it has been widely used. Moreover, the threshold functions are discussed, with some emphasis. Also the paper introduces the noise model, the noise variance estimation and quality metrics about image denoising.2. Research on speckle reduction algorithm, which is based on Bayesian statistical property for medical US Images, has been a heated research problem. The dependence, which still exist among the wavelet coefficients after wavelet transform to images, can be used to estimate the accurate models for probability distribution of wavelet coefficients. On the base of Bayesian statistical property, these models can realize the goal of speckle suppression. This method can considerably enhance the quality of speckle reduction US image. By performing logarithmical transform and followed by redundant wavelet transform, a speckle reduction algorithm, which is based on the local statistical property of wavelet coefficients, is realized for medical ultrasound images. Specifically, the wavelet coefficients of speckle and signal are modeled by two-sided Generalized Nakagami Distribution (GND) and General Gaussian Distribution (GGD) , respectively. Then the thesholding and shrinkage estimators, which are named GNDThresh and GNDShrink, respectively, are deduced under the rule of Bayesian maximum a posteriori (MAP). The experiment results demonstrate that compared with the classical speckle reduction algorithms, the signal-to-noise ratio and the coefficient of correlation have been increased significantly, and the texture has been well preserved.