基于图像处理的汉字识别研究外文翻译资料

Digital image processing

M. Prokop, C.M. Schaefer-Prokop

Department of Diagnostic Radiology I, Medizinische Hochschule Hannover, D-30623 Hannover, Germany

Summary. The image quality of a radiograph is determined by the local contrast, spatial resolution, latitude and the image noise. The goal of digital processing is to improve the visualisation of pathology by optimising these physical parameters. Processing parameters need to be chosen correctly in order to overcome the inverse relationship between contrast and latitude while producing images that retain a conventional appearance. Unsharp mask filtering (UMF) is a simple technique for improving image quality. This technique, however, suffers from serious drawbacks, such as the suppression of pathologic lesions or artifacts that may simulate pathology. Manufacturers have de-veloped different approaches in order to overcome problems and artifacts derived from this technique.

Key words: Digital radiography – Image processing – Quality optimisation

Introduction

The human eye is a major limiting factor for the detection of information contained in a radiograph. There is only a limited number of grey levels that can be differentiated, thus the detection of an abnormality is heavily dependent on its contrast relative to the surrounding structures. The sensitivity to fine detail (high spatial frequencies) is relatively low. In addition, there is a non-linear relationship between perceived grey levels and objective measures of optical density [1].

There are four main factors that determine image quality of a radiograph:

– latitude (minimum to maximum exposure that can be displayed)

– local contrast

– spatial resolution

– image noise.

In conventional radiography, contrast and latitude can-not be varied independently. Both are determined by the gradation curve of the screen/film system. As a result, the optimum screen/film combination depends on the imaging task: wide latitude films are used for chest imaging in order to simultaneously display lungs and mediastinum. However, they will lead to a low contrast in certain areas of the chest. On the other hand, films for skeletal imaging will yield high contrast in the bones but will suffer from an insufficient display of soft tissues or areas with large attenuation differences (e. g. the tho-racolumbar junction).

Digital processing is performed to optimally display the information contained in an image in order to im-prove visualisation of pathology. The goals of image processing are

– to display the full object range (from maximum to minimum attenuation)

– to improve local contrast

– to optimise the spatial resolution of the digital system – to suppress image noise.

Appropriate digital processing is able to independently optimise latitude and contrast and is able to partially compensate for a lower spatial resolution of the digital system. However, the choice of image processing is not trivial: by enhancing one image feature, other features may become suppressed. Problems occur when improper processing parameters are chosen that obscure diagnostically relevant features.

In general, a good display of anatomy is not enough. Processing parameters that are optimised for display of anatomy may hide pathology, may increase image noise, or may lead to artifacts. Modern image processing tools try to produce images that have a conventional appearance but overcome the inverse relationship between contrast and latitude.

Global adjustment of contrast and density

In conventional radiography, the gradation curve of the screen/film system determines overall contrast and density. After having exposed the radiograph, almost no changes are possible. Over or underexposure leads to images that are too light or too dark.

In digital systems, contrast and density can be kept constant, independent of the exposure level, because appropriate adjustments can be made on the basis of the digital raw data. In addition, arbitrary gradation curves can be chosen.

In conventional radiography, the gradation curve of the screen/film system determines overall contrast and den-sity. After having exposed the radiograph, almost no changes are possible. Over or underexposure leads to images that are too light or too dark.

In digital systems, contrast and density can be kept constant, independent of the exposure level, because appropriate adjustments can be made on the basis of the digital raw data. In addition, arbitrary gradation curves can be chosen.

Detection of object range

Storage phosphor systems use sophisticated software in order to detect the collimated areas in which there is relevant diagnostic information. In a second step the histogram of the detected areas is determined. This histo-gram is processed together with other information, such as the organ region that has been imaged, to derive the signal range with diagnostically relevant information. The algorithms are complex and often employ neural network techniques in order to achieve optimum results. Ideally, only the full object range is detected and direct exposure as well as lead collimation are excluded. Given a correct determination of the object range, the resulting images will have comparable optical densities independent of the exposure level and interindividual variations in patient size.

Only the detected signal range is assigned to subsequent processing and display. This is done by rescaling the digital raw data. For chest radiography the available algorithms are very reliable and may only fail in extreme cases. As a result, even under unfavourable imaging conditions, such as in bedside studies, expo-sure errors are virtually eliminated [2]. In skeletal studies, however, more variations occur. According to our own experience, older storage phosphor systems had problems in up to 20 % of the images, depending

剩余内容已隐藏，支付完成后下载完整资料

数字图像处理

M. Prokop, C.M. Schaefer-Prokop

Department of Diagnostic Radiology I, Medizinische Hochschule Hannover, D-30623 Hannover, Germany

总结。通过局部对比度、空间分辨率、纬度和图像噪声等方法，对图像质量进行了分析。数字处理的目的是通过优化这些物理参数来提高病理学的视觉化。为了克服对比度和纬度之间的反比关系，需要正确地选择处理参数，同时产生保留传统的ap-pearance的图像。Unsharp mask滤波(UMF)是一种提高图像质量的简单技术。然而，这种技术却有严重的缺陷，如抑制病理损伤或可能模拟病理的工件。制造商已经开发出不同的方法来克服来自这种技术的问题和工件。

关键词:数字射线照相——图像处理——质量优化。

介绍

人的眼睛是一个主要的限制因素，以检测信息包含在一个x射线图。只有有限数量的灰度可以是不同的，因此对异常的检测严重依赖于其相对于周围结构的对比。对精细细节的敏感性(高空间频率的频率)相对较低。此外，感知到的灰度水平与光学密度的客观度量之间存在非线性关系[1]。

有四个主要的因素决定了一个x射线照片的质量:

-纬度(可显示的最低曝光量)

——当地的对比

——空间分辨率

——图像噪声。

在传统的放射学中，对比度和纬度不能独立地变化。两者都是由屏幕/电影系统的等级曲线决定的。作为一种重新设计，最佳的屏幕/胶片组合依赖于成像任务:宽高纬度薄膜用于胸部成像，以同时显示肺和纵隔。然而，它们会导致胸部某些部位形成较低的对比。另一方面，骨骼成像的薄膜将会在骨骼中产生高的对比，但会受到软组织或具有较大衰减差异的区域(例如tho-racolumbar连接)的不充分显示。

数字处理是为了最优地显示图像中包含的信息，以证明病理学的可视化。图像处理的目标是。

-显示完整的对象范围(从最大值到最小衰减)

-改善当地的对比。

-优化数字系统的空间分辨率，抑制图像噪声。

适当的数字处理能够独立地优化纬度和对比度，并能够部分补偿数字系统的低空间分辨率。然而，图像处理的选择并不简单:通过增强一个图像特征，其他特征可能会被抑制。当选择适当的处理参数时，会出现问题，而这些参数是模糊的诊断相关特性。

一般来说，良好的解剖显示是不够的。为显示解剖而优化的处理参数可能隐藏病理，可能增加图像噪声，或可能导致工件。现代图像处理工具试图生成具有传统外观的图像，但克服了对比度和纬度之间的反比关系。

对比度和密度的全局调整。

在传统的射线照相中，屏幕/胶片系统的渐变曲线决定了整体的对比度和密度。曝光后，几乎没有任何变化。过度曝光或曝光不足会导致图像太亮或太暗。

在数字系统中，对比度和密度可以保持恒定，与曝光水平无关，因为可以在数字原始数据的基础上进行适当的调整。此外，还可以选择任意的渐变曲线。

在传统的射线照相中，屏幕/胶片系统的渐变曲线决定了整体的对比度和密度。曝光后，几乎没有任何变化。过度曝光或曝光不足会导致图像太亮或太暗。

在数字系统中，对比度和密度可以保持恒定，与曝光水平无关，因为可以在数字原始数据的基础上进行适当的调整。此外，还可以选择任意的渐变曲线。

检测的对象范围

存储荧光粉系统使用复杂的软件，以检测有相关诊断信息的准直区域。在第二步，检测区域的组织图确定。这个histo-gram与其他信息一起被处理，例如被成像的器官区域，通过诊断相关的信息来获得信号范围。该算法是复杂的，通常采用神经网络技术来实现最优的重新生成。理想的情况下，只有完整的对象范围被检测到，并且直接暴露和铅准直被排除。给定一个正确的对象范围的确定，得到的图像将具有相对的光学密度，独立于暴露水平和个体差异的患者大小。

只有被检测到的信号范围被分配到低质量的处理和显示。这是通过重新缩放数字原始数据完成的。对于胸部放射学，可用的算法是非常可靠的，在极端情况下可能只会失败。因此，即使在不利的成像条件下，例如在床边的研究中，暴露的错误也几乎被消除[2]。然而，在骨骼研究中，更多的变异发生了。根据我们自己的经验，根据解剖区域的不同，较老的存储荧光粉系统在多达20%的图像上存在问题。现代系统只有在极少数情况下才会失败。

有等级的调整

级配曲线描述了在胶片上的数字像素值与光学密度之间的函数或显示器上的发光度。它的目标是将光学密度调整到目标范围，并将光学密度分配给不同的解剖区域。它与屏幕/胶片上的特性曲线一致。

Fuji-based CR系统的分级fil-tered形象是由曲线类型的GT,曲线的角GA(描述曲线斜率的变化),密度变化GS,曲线中心GC(描述期间保持固定的点在曲线上斜率变化)。该参数系统已被过度终止，即GC是多余的。在Agfa系统中，选择了级配曲线的形式，并利用窗-边技术对对比度和密度进行了交互调整。

在原则上，这两种技术都是等价的:GA与窗口的宽度有关，而GS则与窗口级别相关。分级调整通常是处理过程中的最后一步，在进行空间频率处理后应用。

空间频率处理

空间频率处理改变了数字图像的空间频率组成。上面提到的图像处理的目标可以通过各种各样的数学方法来实现，所有这些都导致了类似的结果。

一般原则

数字图像的各个方面都可以单独分析，可以作为信息处理的基础。举个例子，一个带有肺结节的数字胸片。关于结节，图像包含空间信息(结节的位置)、振幅信息(结节的密度)和空间频率信息(结节的大小和结节边缘的锐度)。每一个都有潜在的诊断相关性。

空间信息主要用于计算机辅助诊断的算法，试图定位病理学。数字图像处理通常集中于振幅和空间频率信息。

巨大的振幅相当于巨大的吸收差异，局部(如钙化结节)或全球(例如肺和纵隔，或股骨和周围软组织之间的)。小幅度的振幅与吸收的小差异相当，要么是由低对比度的物体(例如小的，非钙化的结节)或图像噪声引起的。在一般情况下，可以很容易想象的结构(大的am- pli)可以通过图像处理来抑制，而低对比度的物体(小振幅)则应该是被放大的。因此，当地的对比将会得到改善。

高空间频率(gt; 1周期/mm)描述小图像细节，如肺中的细间隔线或骨骼中的trabeculi。高空间频率也有助于物体轮廓的锐度，例如肺结节的边缘或骨折线的边缘。最后，它们包含了图像噪声的主要组成部分。另一方面，低空间频率(lt; 0.1 cy-cle/mm)包含了关于图像密度的整体变化的信息，例如肺结节与周围环境之间或关节的骨骼和软组织结构之间的双f- fer。由于高空间频率在任何成像系统中都没有得到很好的保护(高空间频率的MTF持续下降)，为了改善小结构的视觉化和物体轮廓的锐度，应该增强它们。另一方面，低空间频率可能会被抑制，因为它们通常包含大量的对比变化(例如肺和纵隔)。结果表明，图像的动态范围缩小，局部对比度得到了证明。

剩余内容已隐藏，支付完成后下载完整资料

资料编号：[22005]，资料为PDF文档或Word文档，PDF文档可免费转换为Word