【作者】 张立伟；

【导师】 李金；

【作者基本信息】 哈尔滨工程大学 ， 生物医学工程， 2008， 硕士

【摘要】 细胞显微图像智能识别是一个较大的难题。白细胞的分割和识别是其中一项非常重要的内容。它的任务是观察和测定血液中各种白细胞的总数、相对比值、形态等,用于判断有无疾病、疾病种类及严重程度。利用自动化仪器代替人工处理,不仅可以大大提高血检工作效率、降低人工劳动强度,也可以使检验更精确。在前人的研究基础上,根据白细胞显微图像的特点,本文给出了以显微镜、彩色CCD及计算机为主体,利用计算机图像分析技术实现白细胞分类的系统。完成白细胞的识别分类,需要以下几个步骤:白细胞显微图像的采集,图像的预处理及分割,图像的特征选择及特征提取,细胞分类识别。首先,根据白细胞的特点,介绍一种显微图像采集装置,对瑞氏染色后的白细胞进行图像采集。其次,对白细胞显微图像进行平滑和锐化等预处理,得到噪声较小的图像。然后,再利用HSI空间中饱和度通道对细胞核的特异性,分割出白细胞细胞核。确定白细胞细胞核的质心,以质心为中心划定一个圆形区域,提取出包含细胞质在内白细胞区域。最后,以白细胞细胞核形状特征和细胞质颜色特征形成特征向量,进行分类识别。形状特征采用对平移、旋转和缩放具有不变性的Zernike矩和HU矩,颜色特征采用整个白细胞非细胞核区的细胞质颜色R、G、B通道均值,并以欧氏距离判断当前颜色归属那种标准颜色。在模式识别中,本论文利用BP神经网络对特征向量进行分类。根据已有的样本完成对BP神经网络中权值和阈值的确定。最后用大量样本进行测试,取得比较满意的效果。本论文在软件设计方面采用面向对象语言开发工具VC++6.0。更多还原

【Abstract】 Intelligent microscopic image Recognition of Blood corpuscle is a main problem. Microscopic image Segmentation and recognition of leucocyte is of real significance. The basic task is to check and calculate the main parameters of leucocytes, such as the quantity, the comparative ratio, the modality, and etc., for diagnosing the presence, type and severity of diseases. This automatic process could not only help improve the efficiency of hemanalysis work and lighten the labor burden but also enhance the precision of examinations.In this thesis, an automatic recognition system of leucocyte is presented using a PC, a microscope and a color CCD camera based on other researcher’s study with the technology in image.In order to recognize the leucocyte, we must take the follow steps: capture of microscopic leucocyte image, pretreatment and segmentation of image, characters selecting and distill and classified recognition of image.Firstly, according to the characters of leucocyte, we introduce an equipment to capture microscopic leucocyte images with Wright’s staining. Secondly, we analyze the feature of leucocyte images and take pretreatments of smoothness and sharpness in order to get the image without noise, and then apply an algorithm of segmentation for nucleus extraction with saturation based on the HSI color space. The leucocyte image including cytoplasm can be extracted approximately in complete with the location of center of gravity for nucleus. Finally, we use the eigenvector with shape feature for nucleus and color feature for cytoplasm to recognize the leucocyte. The shape feature is Zernike moments and HU moments which are invariable for image translation, rotation and zoom. The color feature is the average of cytoplasm color. In pattern recognition module, we selcet BP neural network method to classify the eigenvector. According to the existing example, the weight and limen in BP neural network can be decided. At last, we use a lot of samplers to test recognition system, and get a satisfied result. This research also do plenty of work on software design, we adopt orient-object method to design system and complete all coding in VC++6.0.更多还原