【作者】 聂汉军；

【导师】 沈永增；

【作者基本信息】 浙江工业大学 ， 控制理论与控制工程， 2002， 硕士

【摘要】 有噪图像多分辨率分割中根本问题是噪声的抑制，多分辨率分析充分利用了各分辨率互相独立的信息，具有较好的抑制噪声的能力。Mallat把小波分析与多分辨率分析结合起来，从而使得小波变换多分辨率分析在图像工程中的应用成为人们研究的焦点。 小波分析是近年来发展起来的新的研究领域。它是Fourier分析的突破和进展，是信号和图像处理强有力的工具。本文在对小波分析基本理论研究的基础上，深入探索了小波变换多分辨率分析在有噪图像分割中的应用，并且通过与经典分割方法的比较，体现了小波变换多分辨率分割的优点。 噪声的存在是图像分割的困难所在，本文着重研究了基于小波变换的多分辨率分割中噪声抑制的问题。针对小波变换多分辨率边缘检测中单一阈值难以区分边缘与噪声的问题，本文提出了一种自适应的阈值方法，并改进了边缘链接方法。 中值滤波器是一种去除噪声的非线性处理方法，中值滤波器在边缘保持上具有良好的特性。本文采用模糊Boolean函数提出了一种自适应模糊中值滤波器，并把中值滤波同小波多分辨率分割结合，在抑制噪声的同时较好的保持了边缘。 化学反应中的气液两相图分割存在目标分布紧密、信噪比低的困难，本文利用灰度-小波变换模值的二维分布中，背景、边缘和目标的二维分布之间的欧氏距离较大对图像，进行分割，我们提出了把模糊聚类结合小波变换，在气液两相图分割中取得了成功。更多还原

【Abstract】 The basic principle of multiresolution Segmentation of noise image is suppressing noise. Multiresolution analysis makes use of the independent information in different resolution. Mallat combines wavelet analysis with multiresolution analysis, and multiresolution analysis based on wavelet transform becomes a focal field.Wavelet analysis has been a new field in recent years. It has been accepted as a very important breakthrough of Fourier analysis, and has become a useful tool in image and signal processing. We have study the fundamental theory of Wavelet transform, and then we explore the application of multiresolution analysis based on Wavelet Transform in noise image segmentation. Through comparing between classical image segmentation algorithm and Multiresolution Analysis based on Wavelet Transform, we can know the excellence of Multiresolution Analysis.Suppressing noise is the difficult problem in image segmentation, so we have study suppressing noise in multiresolution segmentation based on Wavelet transform. To discern edge and noise in multiresolution edge detection used single threshold, we put forward a kind of adaptive threshold and improve the method of edge linking.Median filter is a kind of nonlinear filter, and it has good abilities of keeping edge. A new kind of adaptive fuzzy median filter defined by Boolean function is given in this paper, and this method can keeping edge very well and reduce noise at the same time.Gas-liquid image in chemistry reaction has close object and lower signal-to-noise ratio, so gas-liquid image segmentation is very difficult. In 2-D grey-wavelet transform distribution, the Euclidean distance between background, object and edge is very far, so we can carry out image segmentation using this character. We describe a kind of method combing wavelet transform and fuzzy cluster, and succeed in gas-liquid image segmentation.更多还原