基于半分积分投影的车牌字符快速识别算法

A Fast Vehicle License Plate Character Recognition Algorithm Based on the Half Blocked Integral Projection and SVM

提出了一种用于车牌识别的快速字符识别算法。首先利用半积分投影把大小为W×H的二值化字符图像转化为长度为2(W×H)的一维信号,保留了用于图像识别的关键特征,降低了后续算法的复杂度。然后将投影得到的数据进行离散Harr小波变换,抽取第二层小波变换后的低频系数并送入支持向量机进行训练识别。实验结果表明,所提算法可以使车牌字符的总识别率达到97.60%,平均识别时间为16.2ms,有效地提高了识别速度和精度。

A fast character recognition algorithm is proposed for vehicle license plate recognition. Firstly, the character binary image whose size is W x H is transformed into a 1 - dimension signal whose length is 2 （ W ＋ H） by using the half blocked integral projection algorithm, which saves key information of the character to recognition and reduces the complexity of later processing for extracting character. Then the transformed 1 - D signal is decomposed using the discrete Harr wavelet transform in depth of two layers, and the low frequency coefficients of the second layer are extracted as the input parameters of the improved SVM for training and classification. Experiment results show that the proposed method, by which the average recognition rate is about 97.60 % and the average recognition time is about 19 ms, can improve the recognition speed and accuracy efficiently.