摘要
利用SOM神经网络的聚类特性,把图像字符坐标聚成2类拟合成一条直线,求得该直线斜率k,计算倾斜图像的旋转角α,把倾斜图像旋转α,进行水平倾斜校正;在垂直倾斜校正中,使用SOM神经网络聚类拟合直线,计算垂直倾斜角θ,并对水平倾斜校正后的图像进行错切变换,得到最终的校正图像.实验结果表明,该方法能准确地检测出图像的倾斜角,并具有较强的抗干扰性和应用适应性,为图像倾斜校正提供了一个新的有效方法.
A method for license plate tilt correction base on S0M neural network was proposed. According to the feature of S0M clustering, the image character coordinates were divided into two clusters to fit a line and the line slope k was obtained. The rotation angle α was calculated using k and the whole image was rotated by α, namely image horizontal tilt correction was performed. In the vertical tilt correction processing, the line fitting method based on SOM clustering was performed again to compute the vertical tilt angle θ. The rotated image was done, shear transformation and the final correction image was obtained. The experimental results show this algorithm can be easily implemented, the tilt angle can be accurately obtained, and robustness can be offered when dealing with dirty license plate in non-uniform illumination conditions. This method in this paper is more accurate than that of Radon transformation. It also provides a new effective way for image tilt correction.
出处
《长沙交通学院学报》
2007年第4期60-65,共6页
Journal of Changsha Communications University
基金
湖南省教育科学"十一五"规划课题(XJK06CZC078)
关键词
倾斜校正
SOM神经网络
投影法
tilt correction
SOM neural network
projection profile
