【作者】 宋洋；

【导师】 景晓军；

【作者基本信息】 北京邮电大学 ， 军事通信学， 2011， 硕士

【摘要】 随着计算机技术的不断发展,数字图像处理技术越来越得到了更强烈的重视和更广泛的应用。许多新理论和方法也不断涌现出来。在实际应用过程中,由于通信信道和设备、环境等因素干扰,图像信号在产生、传输和记录过程中,经常会受到各种噪声的干扰。数字图像中的原始信号总是和噪声共存的。由于噪声严重地影响了图像的视觉效果,因此在进行边缘检测、图像分割、特征提取、模式识别等工作之前,采用适当的方法减少噪声,即对图像噪声进行滤波,是一项非常重要的预处理步骤。所以使用滤波算法滤除噪声一直以来都是图像处理领域研究的热点。对图像滤波的要求是,既能滤除图像中的噪声又能保持图像的细节。由于噪声和图像细节的混叠,因此在图像滤波中,图像的降噪与细节的保留往往是一对矛盾。图像滤波是图像处理中非常重要的技术环节,至今仍是图像处理领域的研究热点本文首先介绍了数字图像处理以及中值滤波的基本概念,并对基本的图像滤波技术进行了研究。然后本文主要针对中值滤波算法在图像降噪与保护细节上的缺陷,根据不同的方向提出了两种改进算法。一种是根据多级中值滤波提出的基于方向自适应多级中值滤波算法。作者希望通过加强对方向的判断,更好地保护图像的纹理细节。另一种是根据极值中值滤波提出的基于阈值与比值的极值中值滤波算法。作者首先希望通过对极值点更好的分类,保护图像的细节,然后又希望通过对比值的判断使算法更好的适应实际中的噪声,而不仅仅是理论模型。更多还原

【Abstract】 With the continuous development of computer technology, digital image processing technology has increasingly gained more attentions and is more widely applied. Many new theories and methods have been emerged continuously. Image signal is often interfered by many kinds of noises when it is produced, transmitted and recorded in practical application processes because of the communication channels and equipment, environmental factors such as interference. The original signal in digital image is always accompanied by noises The quality of image must be improved by filtering technology before edge detection, image segmentation, feature extraction, pattern recognition and so on. Image filtering is expected to both reduce the noise and keep the delicate details. But it is difficult to design such a good filter because of the overlapping of noise and image details. Image filtering is a very important and hot research field in image processing.Firstly, this thesis discusses some fundamental concepts in digital image processing and median filtering. Secondly, to resolve the contradiction of removing noise and holding image details in median filtering simultaneously, this thesis presents two modified algorithms in different ways. One is direction based adaptive multistage median filtering. The author hopes the details can be protected better through judging the directions of the details. The other is extremum median filtering based on threshold and ratio. The author hopes the details can be kept better by classifying noise more accurately and this algorithm stands the noise brought in our real life better, not only some models in theory.更多还原