基于几何特征的IC芯片字符分割与识别方法

PDF下载HTML

Title:

IC chip character segmentation and recognition method based on geometric features

作者:

郭晓峰^1,2, 王耀南^1,2, 毛建旭^1,2

1. 湖南大学 电气与信息工程学院, 湖南 长沙 410082;
2. 湖南大学 机器人视觉感知与控制技术国家工程实验室, 湖南 长沙 410082

Author(s):

GUO Xiaofeng^1,2, WANG Yaonan^1,2, MAO Jianxu^1,2

1. College of Electrical and Information Engineering, Hu’nan University, Changsha 410082, China;
2. National Engineering Laboratory for Robot Visual Perception and Control Technology, Hu’nan University, Changsha 410082, China

关键词:

IC芯片; 字符分割; 字符识别; 横纵比; 面积比; 几何特征; 最小外接圆; 像素差分

Keywords:

IC chip; character segmentation; character recognition; aspect ratio; area ratio; geometric characteristics; minimum circumscribed circle; pixel difference

分类号:

TP391

DOI:

10.11992/tis.201904028

摘要:

针对IC芯片字符的分割与识别问题，提出了一种基于字符几何特征的分割方法和一种基于字符最小外接圆的归一化与重定位方法，使用基于像素差分的模板匹配完成识别。首先，对芯片图像进行直方图均衡化处理，并利用辅助圆进行中线定位和图像校正，定位得到ROI区域并进行均值二值化处理。随后，对二值化ROI图像进行字符分割，以字符的几何特征作为判断条件，从而完成了对缺陷字符的正确分割。之后，对单字符图像提取最大轮廓，利用其轮廓的最小外接圆进行字符的归一化与重定位。最后，对归一化的字符进行差分识别。通过采集4种芯片样本进行实验，结果表明，该方法能够实现芯片字符的准确分割，对于缺陷字符的分割准确率达90%；能够快速精准地识别芯片字符，单字符平均识别时间为4.6 ms，识别准确率达到99.4%。

Abstract:

To solve the problem of character segmentation and recognition in IC chip, a method based on character geometric features and a normalization and relocation method based on the smallest circumferential circle of characters are proposed. The recognition is accomplished by template matching based on pixel difference. Firstly, the histogram equalization is applied to the chip image, and the auxiliary circle is used to locate the center line and correct the image. The ROI region is located and processed by mean of binarization. Subsequently, the binary ROI region image is segmented into characters, and the geometric features of the characters are used as the judgment conditions, thus the correct segmentation of defective characters is completed. Then, the maximum contour is extracted from the single character image, and the minimum circumscribed circle of the contour is used to normalize and relocate the characters. Finally, the normalized characters are differentially recognized. Four kinds of chip samples are collected for experiments. The results show that the method can achieve accurate segmentation of chip characters, and the accuracy of defective characters is 90%. The average recognition time of single character is 4.6 ms, and the recognition accuracy is 99.4%.