【作者】 段影影；

【导师】 陈武凡； 马建华；

【作者基本信息】 南方医科大学 ， 计算机应用技术， 2010， 硕士

【摘要】 随着现代大型医疗成像设备的飞速发展,新的成像方法和图像处理方法不断涌现,医学图像正成为临床医学研究、诊断和治疗的依据。发展医学图像质量评价方法对于监控和调整医学图像质量、检验和优化医学图像处理算法意义重大。对于医学图像质量的评价,最可靠的方法是主观评价方法。但是医学图像主观评价方法存在代价高、耗时长、实时性差、易受主客观因素影响、不能嵌入实际医学图像处理系统中的缺陷,医学图像质量客观评价方法受到了重视。目前医学图像处理系统中主要采用峰值信噪比(PSNR)来进行医学图像质量的客观评价。由于峰值信噪比没有考虑到像素点间的相关性和人类视觉系统的感知特性,评价结果并不能真实反映图像的视觉感知质量。因此,发展更加符合人类视觉系统特性的医学图像质量评价方法是必要的。本文主要研究医学图像处理技术对医学图像质量影响的评价方法,为医学图像处理技术提供参考。首先介绍了医学图像质量评价方法的研究背景和意义,并介绍了图像质量的两个方面的定义；其次对人类视觉系统进行了简介,介绍了人类视觉系统的基本构造,讨论了与算法有关的人类视觉系统特性和人类视觉的知觉特性。接着对图像质量评价研究的过去和现状进行了概述,详细介绍了图像质量评价的主客观方法,并重点介绍和分析了基于结构相似度的图像质量评价方法的特点和缺点,总结整理了医学图像质量客观评价方法的应用。本文围绕医学图像质量客观评价方法,深入研究人眼视觉特性对图像质量的影响,并着重研究目前较受关注的图像质量评价方法：Zhou Wang等人提出的结构相似度(SSIM)。我们通过对SSIM三个评价因子分解实验,认为SSIM中最重要的结构比较因子难以完整准确的建模人眼的视觉特性。基于此,对图像质量客观评价方法提出了进一步的改进：1.提出一种改进的基于结构相似的图像质量客观评价方法。在改进的算法中充分利用图像梯度与图像边缘和纹理的关系,将图像结构信息重新解释为图像的梯度方向信息,而对比度信息解释为梯度大小,实验结果很好的拟合了主观评价。2.提出了基于梯度加权结构相似的医学图像质量评价方法。充分利用人眼视觉掩盖效应,针对不同的失真类型对于图像不同的内容具有不同的质量影响,提出了一种新的加权策略。新的加权策略利用图像梯度判断图像纹理和失真类型,模拟人眼的视觉掩盖特性,实验结果很好的拟合了主观评价。3.初步建立了用于质量评价的医学图像库,用于图像质量客观评价模型性能评测。从基本方法上说,医学图像质量评价和普通自然图像的质量评价是相同的,但是医学图像的获取方式决定了医学图像与自然图像不同。通用的图像质量评价方法是否适用于医学图像,必须使用医学图像来做进一步的仿真实验。为了更进一步的研究适用于医学图像的客观质量评价方法,我们建立了一套用于质量评价的医学图像库。更多还原

【Abstract】 With the rapid development of the modern medical imaging equipment, both the new imaging methods and image processing methods are constantly emerging. The medical images have become a major reference in clinical research, diagnosis and treatment. The development of medical image quality assessment for monitoring and adjusting the quality of medical image, and testing and optimizing the algorithm of medical image processing is great significance.For all medical image quality assessment, the subjective assessment may be the best one. However, the subjective assessment is too inconvenient, costly and time-consuming for practical usage. Moreover, it is also easily affected by the subjective factors, and hard to be embedded in the medical image processing system. Therefore, how to assess the medical image quality is a very interesting topic.The peak signal-to-noise ratio (PSNR) is used as a major image quality assessment in the medical image processing system. As we all know, PSNR doesn’t take into account correlation between pixels and the perceptual characteristics of human visual system, and the assessment results maybe not reflect the image quality of visual perception. Therefore, it is essential to research some new medical image quality assessment based on the human visual system.In this paper, we study the quality assessment of medical image which to be treated by the medical image processing method, which could be helpful for further medical image processing technique development. Firstly, we introduce the research background and significance, and describe two definitions about image quality. Secondly, we give the brief description of human visual system, and discuss the algorithm that is relative to the characteristics of the human visual system and human visual perception. Then, we outline the image quality assessment, introduce the subjective and objective image quality assessment, and highlights and analysis of the structural similarity based image quality assessment. Moreover, we summarize the application of the objective medical image quality assessment.In this paper, we study in-depth the effects of human visual properties on the image quality, and research SSIM proposed by Zhou Wang etc. We think that it is difficult to modeling exactly characteristics of the human visual system from the structure comparison function in SSIM. Therefore, we propose two new medical image assessments. The main contributions can be summed as follows:1. Improve the objective image quality assessment based on SSIM. In the improved algorithm, we take full advantage of the relation between image gradient, image edge and texture information, re-define the structure comparison function as the image gradient direction information, and the contrast comparison function as the gradient magnitude. The experiment results match greatly with the subjective assessment.2. Proposed a medical image assessment based on gradient-weighted SSIM (GWSSIM). Aiming at different quality effects on different image region and distortion type, we take full advantage of the visual masking effect and proposed a new weighted idea. This new weighted algorithm judges the image texture and distortion type by the image gradient. The experiment results fit well the subjective assessment.3. We construct the medical image database used to evaluate the performance of new objective image assessment methods. For the basic methods, medical image quality assessment is the same as general natural image. But whether the general natural image quality assessment is suitable for medical image must be experimented in simulation by medical image. To study further the better suitable objective medical image quality assessment, we construct the medical image database used to evaluate the performance of new objective image assessment methods.更多还原