中文摘要：

平面几何题目图形识别是智能教育,特别是平面几何题目自动推理重要问题。传统的推理机需要人工录入顶点和其坐标的值,工作量较大且效率低下。针对该问题,提出了平面几何题目图形自动识别的方法,首先对图形采用Otsu方法进行二值化,之后用改进的基于连通域的版面分析的方法进行版面分割,将图像中的字母部分和图形部分相互分离。对分割出来的字母部分,使用模板匹配方法进行识别;对分割出来的图像部分,分别使用Hough变换和修正的Shi-Tomasi角点识别方法进行识别,再使用顶点匹配方法,将字母与对应顶点相联系,选择出需要的顶点信息;最后使用点线匹配的方法,完美地识别出图像中存在的各个三点共线性质。在243道中考题目上测试该方法,与人工标注的顶点坐标相对比,得到完全识别正确率61.5%、邻域像素识别正确率82.1%的结果。实验结果表明,该平面集合题目图形识别算法的正确性较高,基本满足推理机对识别精度的要求。

英文摘要：

Recognizing and understanding of plane geometry is original in intelligent education,especially in automatic reasoning problem of plane geometry. Traditional inference engine requires manual input the coordinates of the vertices,which work is heavy and inefficient. For this problem,an automatic recognization method for plane geometry was proposed. This plane geometry method,firstly,used Otsu method to binarize the image and used improved page analysis method based on connected region to segment the page. Then,it used the template matching method to identify the letter part and the image part,respectively. Hough transformation and Shi-Thomas corner detection method were used to identify the vertices,then vertex matching method was applied to connect characters and their corresponding vertices. The presence of three collinear prosperities could be perfectly identified. This method was tested on 243 senior high school entrance examination topics,the perfect recognization rate is 61. 5% and the accuracy rate of neighborhood pixels is 82. 1%. Experimental results show that the recognization accuracy is acceptable,and it can meet the accuracy requirements of inference engine.