一种新的车牌识别预处理算法

An Efficient Method for Recognizing License Plate

提出了一种八方向Sobel算子模板提取图像边缘的方法，并根据边缘图像中车牌区域纹理特征来切分出车牌区城，然后利用类别方差自动门限法来确定字符色和底色的闷值，以进行字符二值化。经测试，此方法定位准确，二值化效果较好。为后继字符分割和识别做了较好的预处理工作。

Abstract: The percentage of correct recognition of license plate by past methods is, to our best knowledge, still somewhat low; it is particularly low in China. We propose a new method of pretreatment that can make the percentage of correct recognition higher, particularly in China. Our pretreatment method emphasizes the utilization of the texture features of edge image. Accurately extracting edge image is of key importance and we accomplish the accurate extraction with an eight-direction Sobel operator template proposed by Yu et al[4]. We detect the license plate according to its texture features in edge image. Final recognition requires also gray-level threshold of characters and background and we determine such threshold by using method of auto class variance threshold. Figs. 3 through 7 show the successive steps of pretreatment and Fig. 8 shows the final recognition. These figures show preliminarily that our method of recognition is efficient.