摘要
文中提出一种基于神经网络,利用车牌颜色、字符分布特征来提取车牌的算法。与以前的神经网络定位车牌不同的是,本算法是用二值化后每个8-连接对象作为网络的输入。这样可以减少训练样本数目,有针对性地训练噪音。实验证明本算法对于复杂背景的车牌有较好的提取效果,并且有较快的执行速度和较好的鲁棒性。
Presents an algorithm which makes use of information such as color of the car plate and the distribution of the car plate characters based on neural network to extract the car plate. The algorithm uses every 8 - connection objects as the input of the neural network which is different from other neural network location of car plate. The method could reduce the sample used for training the network and training the noise of the picture directly. Experiments show that method is effective and robust for the photo which have complicated backgrounds and meantime execute very fast.
出处
《计算机技术与发展》
2008年第2期38-41,共4页
Computer Technology and Development
关键词
车牌定位
神经网络
颜色空间
灰度共生矩阵
car plate location
neural network
color space
gray level co - occurrence matrix
