摘要
在车牌识别系统中,为提高字符分割和识别的效果,需要对定位后的车牌图像进行倾斜校正。提出了一种基于边缘点投影方差最小的车牌倾斜校正方法。首先将旋转后车牌图像的垂直边缘向水平方向投影;然后导出边缘点投影方差最小时旋转角度的闭合表达式,由此一步确定车牌的倾斜角度;最后通过双线性插值方法校正车牌。给出了实验结果,并与车牌倾斜校正的Hough变换法、旋转投影法和主成分分析法进行了精确度、算法复杂度及运算时间的比较,结果表明所提出的方法精确度高、鲁棒性好、实时性强。通过进一步分析和推广,将车牌倾斜校正方法总结为两种不同类型的方法,并对它们进行了比较。
In order to improve the result of character segmentation and recognition, license plate picture after location should be corrected in the license plate recognition system. A license plate tilt correction method was proposed. First, the vertical edge of plate picture after rotation was projected horizontally. Then a closed formula based on the minimum variance of projected point was found out in order to get the tilt angle. Finally, the plate picture was corrected with bilinear interpolation. The experimental results show highly accurate, robust and real-time. The method was compared with license plate tilt correction methods based on Hough transform, rotation-projection and principal component analysis on precision, complexity and run time. These methods were summarized to two types after further analyzing and generalizing, and they were compared then.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2008年第21期5829-5832,共4页
Journal of System Simulation
关键词
车牌识别
倾斜校正
边缘检测
旋转投影
双线性插值
vehicle license plate recognition
tilt correction
edge detection
rotation-projection
bilinear interpolation
