【作者】 倪臣敏；

【导师】 叶懋冬；

【作者基本信息】 浙江大学 ， 应用数学， 2006， 硕士

【摘要】 图像在采集或者传输过程中,往往会引入不同程度的噪声,这就为后面的边缘检测或者图像分割、形状识别等带来很大的难度,图像去噪便成为图像处理中非常重要的一环。近年来,图像的恢复和去噪引起了广泛地研究兴趣,本文针对脉冲噪声的去除,提出并验证了用于估计脉冲噪声密度的模糊指标,并在此基础上提出了两种脉冲噪声去除算法,通过实验验证了算法优势。 文章首先介绍了数字图像的概念、图像的噪声模型以及经典的噪声滤波算法和评价方法,然后介绍了文章研究的理论背景,包括模糊数学的基础理论和曲线拟合的有关知识两部分。第三、四两章是作者的主要研究成果,也是本文的重点。第三章首先综述了近年来常见的脉冲噪声去除算法;然后基于模糊数学的理论知识以及脉冲噪声的特点,提出了用以判断脉冲噪声强度(即密度)的模糊指标,通过实验统计和曲线拟合获得噪声强度与模糊指标的关系,并验证了其性能;最后引入Prewitt梯度算子,获得梯度阈值限制,改进了中值滤波算法,更好的保持了图像的细节信息,因为有模糊指标判断噪声强度,使得滤波处理的自适应性更好。第四章针对脉冲噪声密度大于50%的噪声图像,首先介绍了一种很好的去噪算法—MMEM算法,然后提出基于序列图像的脉冲噪声去除新算法,利用脉冲噪声的正负脉冲特性,提出点对点的检测算法,充分利用每幅图像的有用信息来恢复受污染的图像,取得了良好的恢复效果。最后第五章总结全文,并对文章中一些问题提出了新的想法和见解。更多还原

【Abstract】 Images are often corrupted by different kinds of noises due to errors generated by noise sensors or communication channels,which makes it more difficult to do some subsequent image processing,such as edge detection,image segmentation and object recognition,so noise detection and removal becomes part and parcel in image processing,and has arrows widely researching interest in recent years. In this article,a fuzzy operator is proposed and verified to evaluate the noise density of impulse noise.Based on this operator,two new filter algorithms are presented for removing impulse noise.Therefore ,better results are expected by experimental comparation.Firstly, the definition of digital images,the image noise model, some traditional filtering algorithms and their evaluating methods are introduced,then the basic mathematical theory used in the following two chapers is given,including the elementary fuzzy mathematics theory and some information about curve fitting.Chapter 3 and chaper 4 are the main fruits of the author,they are the most important parts of the article as well.In chaper 3,at the beginning,the lately impulse noise removal algorithms are summerised;then a fuzzy indicator to evaluate the impulse noise intensity(density) is put forward grounded on fuzzy mathematical theory and the characters of impulse noise,through experimental statistics and curve fitting,the correlation function between noise intensity and fuzzy indicator is obtained and tested;finally,the Priwitt gradient operator is brought in,using restrictive gradient threshold getting from experiment,the median filter algorithm is improved,which leads to better results in saving thin details.As the fuzzy indicator can judge the noise density,the improved median filter is more adaptive.In chapter 4, focus on the highly corrupted (>50%)images by impulse noise,first give a good noise removal algorithm—MMEM algorithm,then a new noise detection algorithm pixel by pixel is proposed using sequential images ,according to the positive and negative character of impulse noise.It searches the useful information in each image sufficiently to restore the corrupted images and reaches better result compared with traditional algorithms.In conclusion,chapter 5 sums up the whole article and gives some new ideas and opinionson some questions in the paper.更多还原