【作者】 张建立；

【导师】 张应中；

【作者基本信息】 大连理工大学 ， 机械设计及理论， 2005， 硕士

【摘要】 工程图纸矢量化是将纸质工程图纸扫描输入计算机后,对所得的扫描图像经过处理和识别,最终重建其中的图形对象的过程,它是图纸复用、自动理解等应用的基础,技术上涉及图形图像处理、人工智能和模式识别等多个领域,具有很高的学术价值和应用价值。本文对工程绘图中扫描图纸的图像处理和识别方法进行了深入的研究。 从位图的基本构成的角度,本文运用最基本的方法,提取图像底层信息,实现对位图的读取以及平移、缩放和旋转等基本操作。仔细分析了ClpenGL环境下图像显示的特点,解决了图像显示中左边或下边不能超出视区的问题,实现了对图像的局部观察。 图像处理方面,针对工程图纸图像中出现的噪声类型,本文采用均值滤波和中值滤波等方法消除噪声,提高了图像的清晰度。然后,采用类间方差判断分析法选择最佳阀值,并以此作为图像二值化的阀值,对灰度图像进行图像分割,将图像所包含的有用信息提取出来,为进一步的识别作准备。 本文采用光栅/矢量混合编辑技术,通过人机交互的方式,直接在光栅图像上对点阵形式的几何图元进行交互拾取,实现对工程图像中直线、圆和圆弧等基本图元的识别。提出了图元的一种拾取算法:首先对图元进行稀疏跟踪,通过图元径向方向上长度的变化获取交点位置,然后将已拾取的光栅图元删除,再根据交点位置填充其它图元的删除缺口。 分析实验结果可知,本文的方法对光栅图像有很好的处理效果和较高的识别率,能有效的抵制了噪声的干扰,特别对图元存在多条复杂交线的情形也取得了良好的效果。更多还原

【Abstract】 Engineering drawing vectorization is the process of processing the scanned image and recognizing the graphic objects and finally, generating the vector format CAD file.Vectorization is the foundation of the activities of reusing old drawing and the automatic interpretation of engineering drawing. Therefore, vectorization has high systematic learning and application value. This paper has carried on the research concentrated on scanned image processing and recognition method in engineering drawings.This paper stands on constitute essentially of files and succeed in reading and basic operations using basic methods such as moving, scaling, rotate etc. After carefully analyzing the characteristics of displaying image in the OpenGL’s entironment, this paper resolves that left or bottom boundary of displaying image can’t overstep the view window, and achieves image part observation.In the research of image processing, this paper clears away noises to lift the image clarity aimed at the noise arisen in engineering drawing by means of mean filtering and medium filtering etc.Then, segments image with the best binarization threshold found by calculating the between-class variance to get useful information and prepare for image recognition.The Raster/Vector Mixed Editing is introduced in this paper. Based on this technology, the raster-formed geometric entities can be vectorized directly from raster image by simple interactive picks. In this paper, we study a new approach to pick raster-formed geometric entities: Firstly, the proposed algorithm sparse traces geometric entities and gets positions of intersecting points by analyzing varieties of the entity’s length in radial direction. Then the picked raster-formed geometric is erased. At last, fills the quadrangle decided by the intersecting points.After analyzing experimental results, we can see the proposed approaches in this paper have a good processing effect and higher correct picking rate for raster-formed geometric entities. Also they have good anti-noise ability, especially for the case where the picked entity is intersected with many other entities.更多还原