基于多特征的车牌字符识别算法  

License Plate Character Recognition Algorithm Based on Multi-feature

作　　者：林炳炎 郭勇[1] LIN Bing-yan(Fujian Polytechnic of Information Technology,FuzhouFujian 350003,China)

机构地区：[1]福建信息职业技术学院,福州350003

出　　处：《长春工程学院学报（自然科学版）》2022年第1期109-114,共6页Journal of Changchun Institute of Technology：Natural Sciences Edition

基　　金：福建省教育厅科研项目(JAT201086);全国高等院校计算机基础教育研究会计算机基础教育教学研究项目(2021-AFCEC-168)。

摘　　要：为了提高车牌字符的识别率,提出了一种基于多特征的字符识别算法。算法首先提取字符的HOG和投影特征参数,把提取出来的特征在SVM中训练。其次,采用多次网格搜索方式,多次训练,寻找SVM的最优参数。最后,把训练好的SVM的模型应用于车牌字符的识别。实验结果表明:字符识别正确率达到99.5%,每个字符的精准率、召回率和F1-score值很高,能很好地识别车牌字符。算法对于字符识别和自动车牌识别系统的应用具有一定的参考价值。In order to improve the recognition rate of license plate characters,a character recognition algorithm based on multi-feature is proposed.Firstly,HOG of the characters features and projection parameters are extracted by algorithm.Secondly,the extracted features are trained in SVM.Then,the optimal parameters of SVM are searched by multiple times of grid searches and training.Finally,the trained SVM model is applied in the license plate character recognition.The experimental results show that the correct rate of character recognition is 99.5%.The accuracy,recall rate and F1-score value of each character are very high.It can recognize license plate characters very well.The algorithm provides a certain value for the application of automatic license plate recognition system.

关 键 词：HOG 投影 网格搜索 SVM 字符识别 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]