摘要
相似字符识别率低会影响整个车牌识别系统的性能,而相似字符之间只有局部特征差异较大,并且相似字符样本数目多少差异较大,目前常用的分类器表现得都不稳定.贝叶斯网络分类器充分利用和综合先验知识与样本信息,无论实验样本和特征数目多少,表现得都很稳定.通过使用几千个测试样本对分类器进行测试,并与其他分类器的识别结果作比较.实验结果表明,在相同的特征下,与AdaBoost分类器、BP神经网络分类器、SVM分类器相比,贝叶斯网络分类器对车牌相似字符的识别有较高的识别率和更高的稳定性.
The low recognition rate of similar characters will affect the performance of the whole car plate recognition system, but similar characters differ from each other mostly in a local part, also the numbers of samples are different, so those classifiers used now have unstable performance. The Bayesian Net- work Classifier has stable performance by making full use of and combining prior knowledge with sample information no matter how many samples and features. Thousands of test samples are used to test Bayesian Network Classifier as well as other classifiers. The experiment result shows that, using the same features, the Bayesian Networks Classifier has a relatively high recognition rate and stable per- formance on similar character recognition compared with AdaBoost classifier, BP Neural network classi- fier and SVM classifier.
出处
《四川大学学报（自然科学版）》
CAS
CSCD
北大核心
2013年第4期775-780,共6页
Journal of Sichuan University(Natural Science Edition)
关键词
车牌识别
相似字符
特征提取
贝叶斯网络分类器
license plate recognition
similar character
feature extraction
bayesian network classifier
