基于垂直投影与卷积相结合的车牌字符分割算法

License plate character segmentation algorithm based on vertical projection and convolution

字符分割是车牌自动识别系统中的关键步骤。提出了一种利用投影和卷积相结合的方法来实现字符分割,先对车牌图像进行中值滤波,再进行竖直差分,减小左右边框的影响;根据车牌的先验知识形成一个可变长的字符模板匹配的脉冲波形,用这个脉冲波形和投影图对应的数组做卷积,定位出第一个字符位置,随着模板长度的不断变大,得出每个字符的坐标。实验表明,该方法能够使得字符分割的结果定位更精准,为字符识别创造了良好的条件。

Character segmentation is a key step in the automatic license plate recognition system. A method is proposed to use projection and convolution to achieve character segmentation： median filtering on the license plate image, then the vertical difference to reduce the effect of left and right margins. According to the prior knowledge of the license plate, pulse waveform that can be stretched and matches character template is formed. The pulse shape is used to convolute with the projection graph corresponding array, which may be taken to locate first character. As the template length becomes larger and larger, the coordinate of each character is given. The experiment demonstrates that the result of the segmentation can make the character segmentation more accurate and create a good condition for character recognition.