收稿日期: 2010-10-09
修回日期: 2011-01-26
网络出版日期: 2011-03-26
基金资助
国家自然科学基金(No.60803058)；江苏省自然科学基金(No.BK2010426, No.BK2008279)；图像处理与图像通信江苏省重点实验室(南京邮电大学)开放基金(No.LBEK2010002)；东南大学校内科研项目基金(No.KJ2010416)资助
Vehicle License Plate Location Based on Machine Learning
Received date: 2010-10-09
Revised date: 2011-01-26
Online published: 2011-03-26
Supported by
国家自然科学基金(No.60803058)；江苏省自然科学基金(No.BK2010426, No.BK2008279)；图像处理与图像通信江苏省重点实验室(南京邮电大学)开放基金(No.LBEK2010002)；东南大学校内科研项目基金(No.KJ2010416)资助
提出了一种基于Adaboost算法与最小同值分割吸收核法角点验证的车牌定位方法. 该方法采用Adaboost算法排除明显的非车牌区域，从而减少车牌候选区域的数量. 在验证阶段，采用SUSAN角点检测方法计算每个经过初筛的候选区域属于车牌区域的概率，并根据该概率值对候选区域进行排序. 最终输出概率值最大的区域作为车牌检测结果. 实验结果表明，使用该方法进行车牌定位无需调整参数也能适应光照变化的应用环境.
关键词： 车牌定位; Adaboost算法; SUSAN角点检测
章品正， 王健弘 . 一种应用机器学习的车牌定位方法[J]. 应用科学学报, 2011 , 29(2) : 147 -152 . DOI: 10.3969/j.issn.0255-8297.2011.02.007
This paper proposes a vehicle license plate locating method based on the Adaboost algorithm and smallest univalue segment assimilating nucleus(SUSAN) corner validation. The Adaboost algorithm is applied for initial classification in order to select target license plate region and reduce the number of candidate areas. SUSAN corner validation is then used to calculate and sort probability of each area belonging to the vehicle
license. The area with the highest probability is taken as the detection result. Experimental results show that the proposed method is robust to different illumination conditions, and the preset parameters produce satisfactory results in different experiments.
/
| 〈 |
|
〉 |
