基于融合LBP和小波矩的肺癌图像精细分类算法研究

发布时间：2018-03-06 12:01

本文选题：医学图像　切入点：特征融合　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：近些年来,科技水平的不断发展为人们生活水平的提高创造了良好的条件。智能手机、相机等先进的图像获取设备正以日新月异的速度在人们的生活中普及。大量的图像随之产生,形形色色的图像数据充斥着人们的生活。图像数据量与日俱增,逐步产生了对这些海量的数据进行有效组织与管理的需求,而最终的目的是实现图像数据的有效分类。强烈的需求吸引着众多的研究学者投入精力研究,逐步衍生了图像分类这一研究领域。与此同时,互联网的快速发展为图像分类领域的发展提供了强有力的支撑,如何在图像分类的过程中降低人工参与的成本,靠计算机来实现分类成为核心的研究内容。从研究的角度讲,计算机应当具有这样的识别与组织能力,在图像分类的整个过程中最大化的节约人力成本。随着研究的不断深入,图像分类技术在很多领域应用并且都取得了可观的研究成果。尤其是在机器视觉等领域,图像分类有着举足轻重的地位。而随着图像分类技术发展的日趋成熟,又逐步衍生出了另一个新的研究领域——图像精细分类领域。在对这一研究方向的逐步探索过程中,研究学者发现图像精细分类技术在食品监控,材料分析,生态环境监控,地质勘测等诸多领域都有非常广泛的实用价值与研究前景。于是计算机视觉及相关领域的研究学者纷纷投入到图像分类技术的研究。与此同时,很多相关的技术如人工智能技术、计算机技术都获得了广泛的发展。这些领域的进步在理论与技术上促进了图像精细分类技术的发展。但是目前来看,将图像精细分类应用在医学图像分类上的研究还比较少。因此本文提出一套完整的算法,实现肺癌图像的精细分类,也就是将肺癌进一步区分为小细胞肺癌、鳞肺癌、腺肺癌、细支气管肺泡癌等。这一分类技术能有力的促进肺癌诊断治疗领域的进一步发展。局部二值模式(LBP,Local Binary Patterns)是一种思想简单高效的纹理特征提取算法。广泛的应用到很多热门的领域,如:人脸识别、医学图像处理、场景的识别分类等。本文主要研究在图像特征提取这一步骤提出基于LBP的融合特征提取算法,并融合进适用于肺癌图像精细分类的技术框架,该框架包括了针对图像精细分类的其它算法步骤。最终将整套的算法应用于肺癌图像的精细分类。本文的研究内容如下:(1)简要介绍图像分类及其相关技术的研究现状,对图像精细分类技术领域的研究工作进行综述,列举图像分类技术在军事、生活等领域的应用。(2)介绍本文框架所依赖的快速模板匹配的图像精细分类技术,主要是图像预处理技术,图像提取技术,特征响应图等。(3)本文介绍了原始的LBP算法以及其相关的经典的改进技术,简要介绍了LBP算法的研究现状以及应用广泛的领域。LBP以往应用在普通医学图像分类技术中时效果较好,但是因为很容易忽略对图像精细分类至关重要的细节信息,且仅有纹理特征不能全面的表示图像信息,因此应用在图像精细分类领域时注定会有劣势。同时,在肺癌图像精细分类这一应用中,以往的普通分类技术无法高效的应用在精细分类领域。针对以上这些提出的问题,本文在无需码本与释文的快速模板匹配框架下,提出了基于LBP和小波矩融合特征的肺癌图像精细分类算法,有效的识别了肺癌图像精细分类所需的细节差异信息,实现肺癌图像的精细分类,是精细分类思想在医学领域应用的一次尝试。(4)本文将改进的LBP融合算法应用于肺癌图像精细分类的系统架构中,在实验部分选用经典的SVM分类器进行后续的分类操作。实验数据库选择LIDC(Lung Image Database Consortium,肺部影像数据库联盟)图像数据集。对比实验与不同特征、不同参数下的融合算法在RBF核函数等环境下进行对比,实验结果表明本文所提算法能有效的应用于肺癌图像的精细分类。
[Abstract]:In recent years, the development of science and technology to improve the living standard of the people has created good conditions. The intelligent mobile phone, camera and other advanced image acquisition devices are popular with the speed of change rapidly in people's lives. A large number of images produced, the image data of every hue filled with people's life. The amount of image data gradually grow with each passing day. Have these massive data for effective organization and management needs, and the final aim is to realize the effective classification of image data. The strong demand attracts many scholars devote research, gradually derived from this research in the field of image classification. At the same time, the rapid development of the Internet provides a strong support for the development of in the field of image classification, how to reduce labor costs of participation in the process of image classification by computer, to achieve classification, become the core of research The contents from the research point of view, the computer should have the identification and organization of such ability in the process of image classification in the maximum saving labor costs. With the deepening of research, image classification technology applied in many fields and has made remarkable achievements. Especially in the field of machine vision, image classification plays an important role. With the development of image classification technology is becoming more and more mature, and gradually derived from another new research field. In the image fine classification of this kind of research gradually explore the process, the researchers found that the material image fine classification technology in food monitoring, analysis, ecological environment monitoring, geological survey and many other fields have practical value and research prospects are very broad. So researchers in computer vision and related fields have put into image Study on the technology. At the same time, a lot of related technologies such as artificial intelligence technology, computer technology has been widely developed. Progress in these areas and promote the development of fine image classification technology in theory and technology. But at present, the research of image fine classification used in medical image classification is less so. This paper presents a complete algorithm to achieve precise image classification of lung cancer, lung cancer is further divided into small cell lung cancer, squamous cell lung cancer, lung adenocarcinoma, bronchioloalveolar carcinoma. This classification technique can effectively promote the further development of the field of diagnosis and treatment of lung cancer. The local two value model (LBP, Local Binary Patterns) is a simple and efficient idea of texture feature extraction algorithm. Widely applied to many popular areas, such as face recognition, medical image processing, scene recognition of this classification. This paper mainly research on image feature extraction this step fusion feature extraction algorithm was proposed based on LBP technology, and integrated into the framework for lung cancer image fine classification, the framework includes the steps of image classification algorithm for other fine. Will eventually set fine classification algorithm is applied to lung cancer images. The research contents of this paper are as follows: (1) briefly introduces the research status of image classification and its related technologies, research work on image classification technology in the field of fine review, cited image classification technology in military applications, and other areas of life. (2) the precise image classification technology this framework relies on fast template matching, is the image pretreatment technology image feature extraction technology, response diagram. (3) this paper introduces the original LBP algorithm and its related classic improved technology, the paper briefly introduces the research status of LBP algorithm As well as the application fields.LBP have been used in ordinary medical image classification technology is better, but because it is easy to overlook the details of image classification is fine, and only the texture features can not fully express the image information in the image, so the fine classification field will be doomed to a disadvantage when used in lung cancer. At the same time, precise image classification the application, application of general classification technology in the past cannot be efficient in fine classification field. Aiming at these problems, this paper without the code template matching and fast text frame, put forward the fine classification algorithm LBP lung cancer images and Wavelet Moment Feature Fusion Based on information, details of the differences of effective recognition the lung image fine classification required for the implementation of the fine classification of lung cancer images, is a fine classification theory in an attempt to medical applications (4) in this paper. The system architecture will be improved LBP fusion algorithm is applied to the lung cancer image fine classification, classification of the subsequent operation in some experiments with classical SVM classifier. Select LIDC (Lung Image Database Consortium experimental database, lung image database alliance) image data set. And the different characteristics of contrast experiment, fusion algorithm under different parameters are compared in the RBF kernel environment, experimental results show that the proposed algorithm can be effectively applied to image classification. The fine lung cancer

【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】