摘要
针对常用投影法单字切分的不足 ,提出基于连通体检测及投影法的字符切分 .首先进行第一次连通体检测 ,选择用于字符群上下边缘线拟合的连通体 .然后用细胞神经网络的方法得到字符群的块状区域 ,用最小二乘法拟合字符群的上下边缘线 .舍弃干扰 ,得到比较“干净”的牌照图象 .对于几何失真较严重的牌照 ,进行几何校正 .对上述处理过的牌照图象 ,进行第二次连通体检测 .对连通体宽度在单个字符范围内的 ,认为连通体的边界为字符的边界 ,对超过单个字符宽度的连通体 ,用投影法重新进行切分 .经测试 ,此方法能较好地去除螺钉和边框的干扰 ,切分位置合理 .
A method of character segmentation is presented in this paper. Connected components are detected the first time to be chosen for fitting upper and lower edge lines. Then massive region where characters are included in image can be obtained using cellular neural networks. And upper and lower edge lines can be fitted using least square. Remove noises whose areas are small and get more cleaner license plate image. For those license plates of geometric distortion, geometry correction is needed. Then connected components are detected the second time. For connected components whose widths are in scope of single character width, borders of connected components are regarded as those of characters. For connected components whose widths are greater than those of single character, segmentation is done another time using projection method. Figures show preliminarily that our method of recognition is efficient.
出处
《小型微型计算机系统》
CSCD
北大核心
2004年第4期564-566,共3页
Journal of Chinese Computer Systems
基金
国家自然科学基金重点项目 ( 60 13 40 10 )资助
关键词
连通体检测
投影法
细胞神经网络
阴影检测器
connected component detection
projection methods
cellular neural networks
shadow detector
