【作者】 吕钢；

【导师】 郑洪源；

【作者基本信息】 南京航空航天大学 ， 计算机应用技术， 2007， 硕士

【摘要】 随着现代科学技术的飞速发展和信息的急剧膨胀,如何对现有的工程图纸文档归档整理和检索,更好的发挥图档资料的作用,已经成为档案部门实现信息现代化管理的主要难题,用户不但要定位图纸,而且需要找到具有相似结构的图纸,工程图纸检索方法的研究将日益重要。本文结合工程图纸的特点,应用图像处理技术,对工程图纸检索技术做了一些深入研究。传统的图纸检索技术都是基于关键字的检索,本文应用图像处理技术,实现对以位图方式存储的工程图纸的检索。基于内容的方式检索工程图纸要求先对扫描后的工程图纸预处理,主要包括二值化,去除噪声,边缘检测,坐标调整,图纸标准化后保存为XML格式,为后续的图纸特征提取做好铺垫。其次,运用图像处理算法对工程图纸提取特征,本文主要提取了密度特征和直方图特征等视觉特征,图元树特征和轮廓区域特征等形象特征。最后,本文用BP神经网络方法对图纸进行分类,在分类的基础上进行多特征组合检索。本文给出了上述方案的系统实现,设计了四种不同类型的图纸的对比实验,对图纸标准化,图元检测,特征提取,分类和检索等算法和实验效果分析评价,并初步应用于图档管理系统中的图纸检索,取得了较好的效果。更多还原

【Abstract】 With the rapid development of the modern technology and sharp expansion of information, it has become the main hard problem of archives department to realize information modernization management, which is how to archive and retrieve the current engineering drawings and document in order to make the drawings and archives effect better. The user not only orients the wanted drawing, but also needs to gather the drawings with the similar structure, what makes it more and more important to do research on the method of engineering drawings retrieval. Combined with the characteristic of engineer drawings, applying image processing technology, this thesis does some deep research on engineer drawings retrieval.Traditional drawings retrieval technology was based on key words while this thesis applies image processing to realize the retrieval of engineering drawings stored in bitmap format. Content-based retrieval method demands that firstly, the scanned engineering drawings are preprocessed, which mainly includes binarization, denoising, edge detection and coordinates regulation, then standardized and saved as XML format, which is prepared for the follow-up feature extraction. Secondly, the engineering drawings features are extracted applying image processing arithmetic. The visual features such as density feature and histogram feature and the imaginative features such as image element tree feature and contour region feature are mainly extracted in this thesis. Finally, the drawings are classified using BP neural network method and on the basis of classification the drawings are retrieved with multi-feature combination method.This thesis presents the implementation of the scheme, designs comparing experiments of four different kinds of drawings, analyses and evaluates on the arithmetic and effect of drawings standardization, image element detection, feature extraction, classification and retrieval etc. We introduce the application in the retrieval of the drawings in the EDM system and get a series of good practical conclusions.更多还原