【作者】 李鹏；

【导师】 石玉英；

【作者基本信息】 华北电力大学(北京) ， 数学， 2022， 硕士

【摘要】 近年来,随着计算机辅助诊断系统和远程医疗在医学中的快速发展,数字图像处理非常关键。图像的边缘涵盖了大部分的图像信息,医学图像边缘检测是进行后续图像处理的基础。因此,研究医学图像边缘检测具有重要的实际意义。本文对医学图像、数字图像处理的应用和前景以及经典的边缘检测算法做了详细的描述和分析。目前边缘检测全变分模型和数学形态学算法是图像边缘检测的主流算法之一,在医学图像处理中具有重要的研究价值。对于医学图像来说,混合噪声的边缘检测是一项具有挑战性的任务,因为混合噪声分布通常没有参数模型。基于医学图像经常存在的高斯和柯西混合噪声,本文提出了一种基于MAP方法的高斯和柯西噪声变分边缘检测模型,采用ADMM方法求解该模型。通过数值仿真实验验证了本文的边缘检测变分模型对于参数的鲁棒性和有效性。最后除了视觉分析,也通过ROC和AUC客观指标验证了该模型均好于其它经典边缘检测算法。基于医学图像成像过程中光源单一以及探测手段的影响,会导致图像噪声分布不均一,往往夹杂多种不同噪声,所以适用于可见光图像的边缘检测算法并不适用于医学图像。本文针对医学图像的特性提出了一种改进的形态学算法,包含以下三种优势。第一,自适应权重赋值。对于多方向结构元素,本文算法根据边缘马氏灰度距离自适应赋值各个方向的权重;对于多尺度多形状结构元素,根据信息熵自适应赋值各个结构元的权重。第二,改进的形态学算子。基于现有算子检测边缘锯齿状,抗噪效果不显著的缺点,本文算法中提出了一种新型抗噪形态学算子。第三,应用于混合噪声彩色医学图像边缘检测。基于现今形态学常应用于灰度图像,为了验证本文算法的鲁棒性,将本文算法应用于四种混合噪声彩色图像进行边缘检测,检测效果良好。最后本文通过视觉直观分析和客观评价指标验证了本文算法均好于其它算法。实验结果表明本文算法提取到的图像边缘完整且清晰,对多种不同混合噪声的抑制和消除也有明显的优势,在医学图像研究中具有很好的应用价值。更多还原

【Abstract】 In recent years,with the rapid development of computer aided diagnosis system and telemedicine in medicine,digital image processing is very important.The edge of the image covers most of the image information and edge detection of medical image is the basis of subsequent image processing.Therefore,the research of medical image edge detection has important practical significance.The application and prospect of medical image,digital image processing and classical edge detection algorithm are described and analyzed in detail.At present,total variation model of edge detection and mathematical morphology algorithm are one of the mainstream algorithms of image edge detection,which have important research value in medical image processing.For medical images,edge detection of mixed noise is a challenging task,because the distribution of mixed noise usually has no parametric model.Based on the mixed gaussian and Cauchy noises in medical images,a MAP based variational edge detection model of Gaussian and Cauchy noises is proposed in this paper,and the ADMM method is used to solve the model.The robustness and effectiveness of the variational model for edge detection are verified by numerical simulation.Finally,in addition to visual analysis,the ROC and AUC objective indicators verify that the model is better than other classical edge detection algorithms.Due to the influence of single light source and detection means in the process of medical image imaging,the image noise distribution will be uneven and often mixed with a variety of different noises.Therefore,the edge detection algorithm applicable to visible images is not applicable to medical images.This paper proposes an improved morphological algorithm based on the characteristics of medical image,which contains the following three advantages.First,adaptive weight assignment.For multi-direction structure elements,the algorithm adaptively assigns the weight of each direction according to the distance of edge Mahalanobis gray scale.For multi-scale and multi-shape structural elements,the weight of each structural element is adaptively assigned according to the information entropy.Second,the improved morphological operator.Based on the disadvantages of the existing operators to detect jagged edges and have insignificant anti-noise effect,a new anti-noise morphological operator is proposed in this algorithm.Third,it is applied to edge detection of color medical image with mixed noise.In order to verify the robustness of the proposed algorithm,the proposed algorithm is applied to four kinds of color images with mixed noise for edge detection,and the detection results are good.Finally,visual analysis and objective evaluation indicators are used to verify that the proposed algorithm is better than other algorithms.Experimental results show that the image edge extracted by the algorithm in this paper is complete and clear,and has obvious advantages in the suppression and elimination of various mixed noises,which has a good application value in medical image research.更多还原