摘要
该文提出了一种新的基于Elman递归神经网络的汽车牌照定位方法。Elman递归神经网络具有上下文层 ,它将隐含层前一时刻的输出反馈到当前时刻的输入 ,这种反馈连接使Elman网络能够检测随时间变化的序列信息。该文的牌照定位方法利用Elman神经网络的反馈连接特性以及牌照区的水平和垂直两个方向的梯度特征不同于图像中的其它区域的特点 ,以一个小窗体 (12× 12 )内图像在两个方向的梯度值 ,对神经网络进行训练 ,然后在同样的梯度图上滑动该小窗体 ,让训练后的神经网络判断小窗体内的区域是否为牌照区的一部分 ,并结合汽车牌照的几何特征来实现牌照定位。
This paper presents an automatic vehicle license plate locating method based on Elman recurrent network. The network has a context layer which feeds the hidden-layer output at previous moment back to the hidden-layer current input. This recurrent connection allows the Elman network to detect the time-varying patterns. In this method, according to the characteristics of the Elman network with feedback and the fact that the gradient features of the plate in the direction of x and y is different from that of other areas in the image, the network is trained. Then a small window (12×12) is used to scan the image and classify each window as plate area or non-plate area with the trained network. And the geometric features of vehicle license plate are also used to locate the plate. Experiments were carried out, and the results show that this method has achieved a very high correct location rate.
出处
《计算机仿真》
CSCD
2005年第2期207-209,共3页
Computer Simulation
关键词
神经网络
牌照定位
梯度
Neural network
Plate location
Gradient
