Abstract: Aiming at the problem of extracting multiple license plates from complex background, an improved two-dimensional discrete wavelet transform was proposed to locate multiple license plates. Firstly, according to the characteristics of wavelet multi-scale decomposition, the two-dimensional discrete wavelet transform was applied to the image to obtain a series of low-frequency wavelet information maps. A linear difference with the original image was made to obtain the detailed features of the license plate characters. Secondly, according to the difference between the characters of license plate and the background in the binary image, the gray level jump was generated, the line position of license plate in the image was roughly located, and the search range of license plate location was narrowed. Finally all license plates were accurately located by using color and shape features, in the small range image of rough location. The research results show that compared with the traditional two-dimensional discrete wavelet transform image, the improved two-dimensional discrete wavelet transform image has nearly doubled the gray level mean value, standard deviation and average gradient, which can effectively obtain the edge information of the original image to narrow the position of multiple license plates. The multi-license plate location method can achieve an accurate positioning rate of 98.96%, with an average time of 328 milliseconds. It can be seen that the proposed method can accurately and quickly locate multiple license plates.

Key words: traffic engineering, multi-license plate location, wavelet transform, image enhancement, gray level jump, line position

摘要： 针对复杂背景提取多个车牌的问题，提出一种改进的二维离散小波变换的多车牌定位方法。根据小波多尺度分解的特性，对图像进行二维离散小波变换，获得一系列小波低频信息图，与原图像做线性差值，获得车牌字符的细节特征；然后根据二值化图像中车牌字符与背景的差异产生灰度跳变，粗定位车牌在图像中的行位置，缩小车牌定位的查找范围；最后，在粗定位的小范围图像中，利用颜色特征和形状特征精确定位所有车牌。研究结果表明：改进的二维离散小波变换图像相比传统的二维离散小波变换图像，灰度均值、标准差和平均梯度提高近一倍，有效获取原图像的边缘信息缩小多车牌位置；多车牌定位方法可以达到98.96%的准确定位率，平均用时328 ms，能够准确、快速定位多车牌。

关键词: 交通工程, 多车牌定位, 小波变换, 图像增强, 灰度跳变, 行定位