摘要
车牌定位是车牌自动识别系统的关键。为了快速准确地进行车牌定位,提出了一种新的RGB颜色空间的车牌定位方法,该方法包括颜色特征提取、特征图像的二值化、形态学连通去噪、车牌候选区域检验4个步骤,同时针对传统颜色特征提取对光照变化敏感的问题,提出了一种不包含亮度信息的颜色特征提取方法;对于车牌候选区域的检验摒弃了常用的易受图像尺寸及图像分辨率影响的区域面积、宽高比、矩形度等几何特征,而是采用车牌字符数及字符排列的规则度作为判定的依据,并由此设计了一种车牌字符规则度的计算方法,用来检验车牌候选区域。通过对包含不同尺寸、不同光照条件的605幅图像进行车牌定位的实验表明,成功率超过96%,可见该颜色特征提取方法对光照变化不敏感,该车牌检验方法可适用于各种不同尺寸图像的车牌定位。
License plate（LP） location is key to vehicle license plate auto recognition system. In this paper, we propose a novel LP location method on RGB color space, which includes four steps： color feature extraction, feature images binaryzation, area filling and de-noising with morphology processing, plate candidate regions verification. Traditional color feature extraction is sensitive to iluminition based on which, we propose a new color feature extraction method that doesnt include the intensity component. For plate candidate regions verification, we discards some geometry features such as the area, length/width ratio, rectangle degree etc, those of which are widely used but sensitive to image size, and we utilizese the special characteristic of LP on character number and character arrangement uniformity to verify the candidate regions. 605 images with various size and illumination condition were tested, success rate of LP location was over 96%, which shows the color feature extraction method is robust to illumination change, and the plate verification is suitable for various image size.
出处
《中国图象图形学报》
CSCD
北大核心
2010年第11期1623-1628,共6页
Journal of Image and Graphics
关键词
车牌定位
RGB颜色空间
车牌字符规则度
数学形态学
license plate location RGB color space character uniformity of license palte morphology
