全文下载：2012060624
陈为1,2， 路长厚1， 陈静2
(1.山东大学 机械工程学院，山东 济南250061；2．青岛科技大学 自动化与电子工程学院，山东 青岛 266042)
摘要： 针对压印字符表面的立体特征，提出了一种基于表面三维形状信息的金属铭牌压印凸凹字符图像分割的方法。使用一种简化的四光源立体技术获得铭牌图像各点的法向量后，基于图论聚类方法，实现了压印凸凹字符的分割。同时，在传统的开环图像处理中引入了闭环反馈控制，获得了分割参数的优化。较传统的基于二维图像分割的方法，本方法的分割结果更令人满意，具有很好的鲁棒性和自适应性，提高了字符的识别率。
关键词： 压印凸凹字符； 图像分割； 光度立体技术； 图论聚类； 闭环反馈控制
中图分类号： TP394.41文献标志码： A
Segmentation of Embossed Characters Pressed on Metallic Label
CHEN Wei1,2, LU Chang-hou1, CHEN Jing2
(1.School of Mechanical Engineering,ShandongUniversity,Jinan250061,China;
2.SchoolofAutomationand Electronic Engineering,QingdaoUniversityof Science and Technology,Qingdao266042,China)
Abstract: A new method is proposed for segmenting embossed characters pressed on metallic label based on surface 3D shape information. After the 3D surface normals of the metallic label are estimated through a simplified 4-source photometric stereo technique, the embossed characters are segmented by grouping their pixels according to their estimated normals through graph-based clustering. The closed-loop control is introduced to overcome the problems of traditional open-loop image processing, which achieved the optimization of segmentation parameter. Compare with the segmentation algorithms based on 2D information, the new algorithm takes advantage of the 3D surface characteristics and obtain a satisfactory segmentation results.The algorithm is robust and self adaptive, which can improve the characters recognition rate.
Key words: pressed embossed characters; image segmentation; photometric stereo; graph-based clustering; closed-loop control
收稿日期:2012-06-06
基金项目： 山东省优秀中青年科学家科研奖励基金项目（BS2010ZZ007).
作者简介: 陈为(1974—), 男, 副教授.
