1、 第 I 页 摘 要 为了对车牌字符的识别，本文将 BP 神经网络应用于汽车车牌的自动识别，在车牌图像进行预处理后的基础上，重点讨论了用 BP 神经网络方法对车牌照字符的识别。 首先将训练样本做图像预处理，对车牌上的字符进行分割，得到单个字符。对大小不一的字符做归一化后，对字符进行特征提取，把长为 15，宽为 25 的归一化后的图像中的字符信息提取出来，图像中白点置为 0，图像中的黑点置为 1，这样就得到了 15 25 的特征向量，这个特征向量记录的就是字符的特征。把这个特征向量送到 BP 网络中进行训练，得到了训练好的权值，把他保存到“ win.dat”和“ whi.dat”中。然后打开要识
2、别的图片（即车牌），对 图像进行预处理后就可以识别了。识别率也在 90%以上，表明该方法的有效性。 关键字： 车牌识别； LPR； 字符识别；特征提取； BP 神经网络； 第 II 页 Abstract For the discernment to the number plate character, this text applies BP neural network to the automatic discernment of the automobile number plate, on the basis that the number plate picture goes on
3、 in advance treated , is it use BP neural network method to car discernment , license plate of character to discuss especially. Will train samples to do the pretreatment of the picture at first, character in number plate cut apart, get the individual character. After making normalization to the char
4、acter not of uniform size, drew the characteristic to the character 15, wide to draw out for character information of 25 picture behind the normalization, picture white point it puts to be 0, black point of picture is it as 1 , receive 15* 25 characteristic vector quantity like this to put, what the
5、 vector quantity of this characteristic is written down is the characteristic of the character . Send the characteristic vector quantity BP network train, get good right value of training, keep him in win.dat and whi.dat. Open picture (namely number plate) discerned to want, go on to picture in advance treated to can discern. The discerning rate is above 90% too； show the validity of this method. Key word： The number plate discerning； The character discerning； LPR； The characteristic is drawn； BP neural network；
