车牌字符的类欧几里得距离特征提取与分析

Feature Extracting of License Plate Character Based on Euclidean-Like Distance Transform

在自动车牌识别系统的设计中,为了实现汉字、字母、数字的快速准确识别,需要一种能够同时提取车牌字符中汉字、字母、数字的统一分类特征,研究利用脉冲耦合神经网络的自动波扩散特性,简化模型参数,提取每个汉字、字母、数字的类欧几里得距离图像,将其作为分类特征,采用最小方差方法测试了特征的有效性。针对不同省市的民用车牌的实验测试结果表明,简化脉冲耦合神经网络模型的卷积核矩阵为5×5时所得的类欧几里得距离图像作为特征向量时,测试图像与标准图像之间的所得方差值最小,能够实现正确匹配。简化脉冲耦合神经网络的类欧几里得距离变换优于传统基于神经网络的特征提取方法,充分利用了图像边缘的形状信息,能够有效地进行分类识别,为汉字、字母、数字全网互联提供了一种统一标准的特征提取方法。

In order to recognize the Chinese characters, letters, and numbers quickly and accurately in automatic number plate recognition system, a unified classification feature is needed. So automatic wave diffusion characteristic of Pulse Coupled Neural Networks （PCNN） was adopted to extract feature vectors form the Euclidean like distance images of each Chinese character, letters, and numbers. Parameters of PCNN wer simplified simultaneously. Then a minimum variance method was utilized to test the validity of this feature vectors. Experimental results based on civil ian license plate from different provinces show that the Euclidean like distance images based on simplified pulse coupled neural network model with 5 x 5 convolution kernel matrix can obtain the smallest square difference between the test images and standard image. Euclidean like distance transform takes full advantage of the shape of the edge of the image, acquire better results than the traditional artificial network based features extraction method for classi fication of the characters.