基于方向轮廓的小波分解车牌字符识别方法

字符识别是车牌识别的技术核心。本文提出了一种快速的车牌字符识别方法。首先提取字符的外轮廓。根据小波分解变换的性质,对字符轮廓进行多分辩分解,得到低分辨率下字符轮廓的近似,通过计算待识别字符与参考字符在多尺度下的轮廓之间的相似度,判别待识别字符与参考字符的匹配程度,最终确定匹配字符。本文对字符匹配算法进行了改进,加快了匹配速度,提高了匹配的可信度。利用本文算法,对1200幅车牌字符图像的识别,准确率达到95.4%,单字符识别平均耗时约为10ms。

A Method of Vehicle Plate Character Recognition Under Multi-scale Directional Contours with Wavelet Transform

Character recognition is key of license plate recognition . A novel method to recognize the character is proposed in this pa-per. First, getting the directional contour of the character, the approximately contour was picked by decomposing with wavelet trans-form. Detecting the similitude of multi-scale contours between the character waiting for recognizing and the referenced character by calculating their comparability ,finally, the matched character was got. The match algorithm was modified in the paper which speedup the match and enhance the reliability. Experiments had been conducted, 1200 vehicle character images were tested, the accuracy was 95.4% and the average consuming time was 10ms.