基于小波的车牌汉字特征提取

车牌识别系统是智能交通不可或缺的一部分.在车牌识别系统中,车牌首位汉字的特征提取和识别是一个难点.为此,将小波的多分辨率特性应用于车牌汉字特征的提取,提出了一种直接从灰度图象提取车牌汉字特征的提取方法.该方法首先提取图象的小波矩和基于小波分解的区域密度特征,然后以识别率好坏为最优依据,进行特征分量排序和选择,最后把特征矢量送入BP神经网络进行字符识别.该方法避免了一些传统汉字特征提取方法需要对图象进行二值化操作而造成的汉字字符结构信息丢失.提取的特征有效地反映了车牌汉字的局部和全局特征.实验结果表明该特征提取方法可以得到比较好的识别效果.

The Car Plate Chinese Character Feature Extraction Based on Wavelet

The car plate recognition system is an indispensable part of intelligent traffic system. In car plate recognition system, one of the most difficult problems is the feature and recognition of car plate Chinese character. The multiresolution property of wavelet is applied to extract car plate Chinese character feature extraction, and a method that directly extracts feature from gray scale image is proposed. The method firstly extracts two kinds of features: one is wavelet moments in different multiresolution, and the other is the zoning density of wavelet-decomposed image. Then, two kinds of features are combined and a feature selection algorithm based recognition rate is presented to choose better features. Finally, the selected features are as the input of BP neural network, which is adopted to recognize car plate Chinese character. As a result, the proposed method avoids the binary operation used in some traditional Chinese character feature extractions that will seriously destroy the Chinese character structure. Furthermore, the extracted features can describe the local and global property of the character. Compared with some other feature extraction methods, the proposed method can achieve better recognition performance.