摘要
运用支持向量机对车牌字符进行识别,解决了由于图像受客观条件的影响、样本数量不是很大等原因导致的识别率不高的问题.主要针对车牌字符中的数字进行实验,选取了15组数字样本,8组进行训练,7组进行测试,采用交叉验证的思想对SVM进行参数C与g的寻优,并选择合适的核函数,对样本进行训练和预测,对于某些数字的识别率可达到100%,并在相同的训练集和测试集下与BP网络的识别效果进行对比.实验结果表明,SVM在训练样本较少且无字符特征提取的情况下具有很好的识别率,并且有很好的分类推广能力.
The application of SVM is presented in vehicle licence recognition, and to avoid the problem of the low recognition rate dependency on feature extraction and a not enough sample. This article mainly aims at the numbers of the characters, select 15 number samples, 8 groups for the training, 7 foe the test, use cross validation to find the best C and g, select appropriate kernel function, after training and testing, some digital recognition rate can reach 100%, and compared with BP which using the same data. Experiment results show that： SVM has high recognition rate which can be extended greatly.
出处
《数学的实践与认识》
CSCD
北大核心
2012年第23期138-143,共6页
Mathematics in Practice and Theory
基金
中国第48批博士后科学研究基金(20100481307)
关键词
车牌数字识别
支持向量机
核函数
交叉验证
license plate number recognition
SVM
kernel function
cross validation
