基于小波分析的倾斜车牌图像字符识别仿真
|
摘要
针对传统倾斜车牌图像字符识别率较低、识别精度不高的问题,提出一种基于小波分析的倾斜车牌图像字符识别方法。首先对原始倾斜车牌图像转换成灰度图像预处理,获得图像像素点,利用小波矩对预处理后的车牌图像像素点提取字符特征,运用直线拟合方程矫正字符倾斜角度目标图像,然后通过主分量分析对提取的字符特征进行选择,并降低字符特征维数,获得特征向量,并输入到BP神经网络,通过对该网络权值初始化,分析输入层与输出层之间的线性映射,并通过线性映射关系完成字符识别。实验结果表明,所提方法能有效提高倾斜车牌图像字符识别率。
A method to recognize characters in slanted license plate image based on wavelet analysis was proposed. First of all, this method converted the original slanted license plate image into gray image for pretreatment, and then the pixel points of image were obtained. After that, our method used wavelet moment to extract the character features from the pixels in license plate image after the pretreatment. Moreover, this method used straight-line fitting equation to correct the target image of inclination angle of character. Meanwhile, we selected the extracted character features by principal component analysis and reduced the character feature dimension to get feature vector which was inputted to BP neural network. By initializing this network weight, we analyzed the linear mapping between the input layer and the output layer. Thus, we completed the character recognition through the linear mapping relationship. Simulation results show that the proposed method can effectively improve the recognition rate for characters in slanted license plate image.
A method to recognize characters in slanted license plate image based on wavelet analysis was proposed. First of all, this method converted the original slanted license plate image into gray image for pretreatment, and then the pixel points of image were obtained. After that, our method used wavelet moment to extract the character features from the pixels in license plate image after the pretreatment. Moreover, this method used straight-line fitting equation to correct the target image of inclination angle of character. Meanwhile, we selected the extracted character features by principal component analysis and reduced the character feature dimension to get feature vector which was inputted to BP neural network. By initializing this network weight, we analyzed the linear mapping between the input layer and the output layer. Thus, we completed the character recognition through the linear mapping relationship. Simulation results show that the proposed method can effectively improve the recognition rate for characters in slanted license plate image.
引文
[1] 董峻妃,郑伯川,杨泽静.基于卷积神经网络的车牌字符识别[J].计算机应用,2017,37(7):2014-2018.
[2] 贾文其,李明,朱美强,等.基于栈式降噪自编码神经网络的车牌字符识别[J].计算机工程与设计,2016,37(3):751-756.
[3] 高强,刘斌.基于提升小波的矩不变量车牌字符识别方法[J].量子电子学报,2016,33(6):662-670.
[4] 王钦民,李宽,杨灿群.一种基于分类器投票的车牌定位方法[J].计算机工程与科学,2016,38(6):1200-1206.
[5] 梁大宽,韩晓明.基于融合Hu矩和区域矩特征的多车牌定位[J].计算机工程与设计,2016,37(11):3097-3101.
[6] 韦玉科,谢嘉乐,吴齐云.基于改进的LM-BP神经网络的车牌字符识别研究[J].测控技术,2016,35(2):48-51.
[7] 宋春静,丁香乾,郭乙运.基于深度信念网络的集装箱字符识别方法[J].计算机工程与设计,2016,37(3):742-745.
[8] 黄艳丹,李军,金星,等.气枪信号触发台站的自动快速识别[J].地震工程学报,2017,39(2):234-239.
[9] 吕颢,刘峰,干宗良,等.基于字典学习的模糊车牌中文字符识别[J].计算机技术与发展,2017,27(10):75-78.
[10] 王晶.基于深度学习神经网络的车牌字符识别技术的研究[J].工业控制计算机,2017,30(3):51-52.
[2] 贾文其,李明,朱美强,等.基于栈式降噪自编码神经网络的车牌字符识别[J].计算机工程与设计,2016,37(3):751-756.
[3] 高强,刘斌.基于提升小波的矩不变量车牌字符识别方法[J].量子电子学报,2016,33(6):662-670.
[4] 王钦民,李宽,杨灿群.一种基于分类器投票的车牌定位方法[J].计算机工程与科学,2016,38(6):1200-1206.
[5] 梁大宽,韩晓明.基于融合Hu矩和区域矩特征的多车牌定位[J].计算机工程与设计,2016,37(11):3097-3101.
[6] 韦玉科,谢嘉乐,吴齐云.基于改进的LM-BP神经网络的车牌字符识别研究[J].测控技术,2016,35(2):48-51.
[7] 宋春静,丁香乾,郭乙运.基于深度信念网络的集装箱字符识别方法[J].计算机工程与设计,2016,37(3):742-745.
[8] 黄艳丹,李军,金星,等.气枪信号触发台站的自动快速识别[J].地震工程学报,2017,39(2):234-239.
[9] 吕颢,刘峰,干宗良,等.基于字典学习的模糊车牌中文字符识别[J].计算机技术与发展,2017,27(10):75-78.
[10] 王晶.基于深度学习神经网络的车牌字符识别技术的研究[J].工业控制计算机,2017,30(3):51-52.