工业X射线图像增强算法研究

发布时间：2018-02-13 02:09

本文关键词： 射线图像增强各向异性扩散变分法对比度场非局部均值 Retinex　出处：《中北大学》2016年博士论文　论文类型：学位论文

【摘要】：随着各种相关需求的不断增加,X射线检测技术已得到了广泛的应用,在医疗诊断、安全检查、航空航天、工业检测等领域均发挥着重要的作用,已然成为我们生产生活中不可或缺的部分。然而,在X射线检测系统成像过程中,由于工件结构的复杂性以及成像过程中电气噪声、射线噪声等各种不利因素的影响,导致工业射线图像的整体质量较差,具体表现为图像对比度低、结构细节模糊等,进而影响对工件结构的分析和判断。因此,在实际应用中,为了使射线图像能够满足具体要求,有必要改善X射线图像的质量,以利于对工件结构的分析与判断。而图像增强则可以依据具体应用要求突出图像中细节特征、提高图像对比度,从而改善图像视觉效果。因此,研究优质高效的X射线图像增强算法具有特别重要的理论意义与实际价值。鉴于此,本文在深入分析、研究国内外相关图像增强算法的现状、分析总结了其在实际应用中存在的问题与不足的基础上,对X射线图像增强算法进行了比较深入的研究,提出了一些新的方法和思路,取得了较好的效果。本文的主要工作如下:(1)在研究偏微分方程中各向异性扩散模型的基础上,提出了一种基于局部方差的自适应各向异性图像增强算法。该算法采用局部方差检测图像的边缘特征,然后设定随迭代次数增加而减小的方差阈值,对图像特征进行分类,在含噪的平坦区域,采用正向扩散,有效去除噪声;在边缘细节区域,采用逆向扩散,增强图像中的特征。实验结果表明该算法克服了传统Laplace增强算法对噪声敏感的不足,有效地增强了图像细节。(2)针对图像梯度对噪声敏感的缺点,提出了一种改进的基于变分方法的对比度场增强算法。该方法利用图像的差分曲率来替代梯度,放大系数随着差分曲率的增加而单调递减。以差分曲率作为自变量的放大系数函数考虑了更多的邻域像素,不仅包括梯度方向上的四个邻域像素,还包括对角线方向上的四个邻域像素,从而克服了图像梯度对噪声敏感的缺点。将该方法用于标准测试图像和X-射线图像,实验结果表明,本章算法增强后的图像边缘更加清晰,细节更加明显,噪声也更少,表现出优越的细节保护能力和噪声抑制能力。(3)在深入分析非局部均值滤波算法原理及特点的基础上,提出了一种基于非局部差值信息的图像增强算法。新算法结合经典的高提升图像锐化算法,引入了非局部均值思想用以计算邻域内像素参与增强操作的权重值,邻域大小可变。同时,中心点像素的权重值可调,从而体现出高提升的效果。因而,新算法继承了非局部均值滤波算法与高提升图像增强算法的优点。通过实验深入分析研究了相关参数对算法性能的影响,并总结了参数选择的原则。此外,针对经典测试图像与实际采集的工业射线图像的增强,从与同类算法性能比较方面进行了实验验证。实验结果表明,经所提算法增强后的图像细节信息更加突出,同时还保持了较高的峰值信噪比,获得了较满意的效果。(4)在深入分析Retinex传统算法的基础上,研究了一种基于人眼视觉特性的改进算法。文中讨论了单尺度算法、多尺度算法对图像增强的影响,选取不同的参数值,图像呈现不同的增强效果。算法采用S函数模型将非线性运算引入到图像增强中,S函数模型更能符合人眼特性,并且可以克服对数处理模型拉伸超出图像显示范围的问题,算法还从保护图像亮度和降低射线噪声对图像的影响两个方面做了改进。实验结果表明该方法不仅对普通灰度图像具有增强效果,还对具有低对比度的射线图像有良好的增强效果,可以满足射线图像后续处理分析的要求。
[Abstract]:With the increasing demand of various related, X ray detection technology has been widely used in medical diagnosis, safety inspection, aerospace, industrial detection plays an important role, has become an indispensable part in our production and daily life. However, in the X ray detection system in the imaging process, due to electrical noise the complexity of structure and the effect of X-ray imaging process, noise and other kinds of unfavorable factors, leading to the overall poor quality of radiographic images, the specific performance of low image contrast, details of the structure fuzzy, and then affect the analysis and judgment of the structure. Therefore, in practice, in order to make the X-ray image to meet the specific it is necessary to improve the quality requirements, X ray image, in order to facilitate the analysis and judgment of the workpiece structure. And image enhancement can be based on the specific application requirements of outstanding image The detail characteristic, improve image contrast, so as to improve the visual effect. Therefore, the research of X ray image quality enhancement algorithm is of great theoretical and practical value. In view of this, based on the thorough analysis, the present research situation of enhancement algorithm of image at home and abroad, analyzes the existing in the practical application the problems and shortcomings on the X-ray image enhancement algorithm for X are studied, and puts forward some new methods and ideas, and achieved good results. The main work of this paper are as follows: (1) based anisotropic diffusion model in the study of partial differential equations, this paper proposed an adaptive anisotropic the specific image enhancement algorithm based on local variance. The algorithm uses edge detection image local variance, and then set as the number of iterations decreases the variance threshold of image features The classification, on a flat area with noise, the use of forward diffusion, effectively remove the noise in the edge region; by reverse diffusion, enhanced features in the image. The experimental results show that this algorithm overcomes the deficiency of algorithm is sensitive to noise enhancement of traditional Laplace, effectively enhance the image details. (2) aiming at the disadvantages of image gradient noise sensitive, proposes a variational method of contrast enhancement algorithm based on improved image. The method uses the differential curvature instead of gradient, with increasing differential amplification coefficient of curvature decreases. The differential amplification coefficient of curvature as the function arguments considering the neighborhood pixels more, not only including four pixel neighborhood gradient direction, also includes four neighborhood pixel diagonal direction, thereby overcoming the image gradient sensitivity to noise. The method used in the standard test images and X- The X-ray images, the experimental results show that the algorithm in this chapter after the image edge enhancement is more clear, the details are more obvious, less noise, showing details of noise protection ability and superior suppression ability. (3) based on the in-depth analysis of the principle and characteristics of non local mean filtering algorithm, this paper proposed an image non local based on the difference information enhancement algorithm. The new algorithm combines high image sharpening algorithm, introduced the idea of non local means used in the computation of neighborhood pixels involved in the enhancement of the operations of the weight value, the variable neighborhood size. At the same time, the weight of the center pixel value is adjustable, which reflects the high lift effect. Therefore, the new algorithm the inheritance of the non local mean filtering algorithm and image enhancement algorithm has the advantages of high lift. Through the in-depth analysis of the relevant parameters on the performance of the algorithm, and summarizes the principle for choosing the parameters in addition, To enhance the industrial radiographic image classic test image and actual acquisition, verified from the performance comparison with other algorithms. The experimental results show that the proposed algorithm enhances the image details more prominent, while maintaining the high peak signal-to-noise ratio, obtained satisfactory results. (4) based on a thorough analysis of the traditional Retinex algorithm, an improved algorithm is studied based on human visual characteristics. This paper discusses the single scale algorithm, multi-scale algorithm of image enhancement, different parameter values, the image shows various enhancement effect. The algorithm uses the S function to model the nonlinear operations into the image enhanced, S function model can accord with human characteristics, and it can overcome the logarithmic model stretched beyond the scope of the problem of image display, image brightness and protection algorithm from reducing shot noise on the line The two aspects of image influence have been improved. The experimental results show that this method not only enhances the effect of ordinary grayscale images, but also enhances the low contrast radiographic images, which can meet the requirements of subsequent processing and analysis of radiographic images.

【学位授予单位】：中北大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】