基于低质量图片的两级车牌字符识别算法  

A TWO-STAGE LICENSE PLATE CHARACTERS RECOGNITION METHOD BASED ON POOR QUALITY IMAGES

作　　者：万燕[1] 刘伟[1] 

机构地区：[1]东华大学计算机科学与技术学院,上海201620

出　　处：《计算机应用与软件》2012年第11期281-284,共4页Computer Applications and Software

摘　　要：目前大多数道口的视频监控系统或图像采集设备都采用普通摄像头,车辆图片质量不高,容易受到光照不均、运动模糊及摄像角度的影响,图片中车牌字符小,字符混淆程度严重,大大降低了车牌字符的自动识别率。针对低质量车牌图片中车牌字符识别率低的问题,提出一种结合支持向量机(SVM)和字符局部特征提取的两级组合分类识别架构。第一级分类器采用核主成分分析(KPCA)对车牌字符进行特征提取,并利用SVM进行分类。如果是易混淆字符,则进入第二级分类器,针对易混淆字符的局部特征设计不同的分类方法加以区分,进而得到最终的识别结果。实验表明该两阶段分类方法能够在各种复杂场景下针对低质量图片达到较高的车牌字符识别率。Recently,most video monitoring systems or image acquisition devices used at traffic intersections are the ordinary cameras,the quality of vehicle images taken by them is poor,and it is susceptible to the impacts of uneven illumination,motion blur and shooting angles.Small in size and severe in confusion the characters images of the license plate are have significantly reduced the automatic recognition rate of plate characters.In this paper,a two-stage composite classification recognition architecture is proposed in light of the problem of low characters recognition rate on poor quality images of license plate,it combines the support vector machine（SVM） with characters＇ local feature extraction.The first stage classifier uses kernel principal component analysis（KPCA） to extract the features of plate characters,and uses SVM to classify them.If the character is the one easily confused,then the second stage classifier works.Aiming at the local features of the characters prone to confusion,different classification methods are designed for distinguishing,and final recognition result is to be achieved then.Experimental results show that the presented two-stage classification method is capable of reaching higher recognition rate for characters on poor quality images of license plate in various complicated scenes.

关 键 词：车牌字符识别 KPCA SVM 局部特征 两级分类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]