摘要
提出一种彩色图像车牌检测方法,主要包括三个步骤,首先对车牌图像执行多尺度区域生长分割,然后在分割结果中进行车牌区域的粗定位,最后对粗定位提取的车牌区域作精确定位分析。区域生长分割方法可以产生封闭的区域,容易提取其形状信息,而长矩形是车牌目标一个显著特征,因此可以通过形状信息检测车牌。在图像分割中,采取多尺度的分割方法,生成多个不同细节程度的分割结果。在每个分割图像中搜索具有车牌外形特征的图像区域,初步检测出候选车牌目标。根据车牌字符的大小和笔画特征,对候选车牌区域进行精确判别分析,得到最终的车牌检测结果。实验部分验证了车牌检测的有效性。
A license plate detection method was proposed. The method includes three steps：multi-scale image segmentation, rough location of license plate, and analysis of extracted image region. As region-growing image segmentation produces closed image regions, the shape information of all the regions is extracted from them. One of the important characteristics of a license plate is its rectangular shape,so this paper used the shape characteristics to detect license plate in the first place. To produce image regions, we used multi-scale segmentation method, by which the color image was segmented into multiple resulting images with different detail levels. The search for license plate like image regions in these resulting images was done to detect candidate license plates. Then the candidate image regions were examined according to the stroke features of license plate characters. The validity of the proposed detection method is shown in our experiment.
出处
《计算机科学》
CSCD
北大核心
2009年第4期261-263,276,共4页
Computer Science
基金
国家自然科学基金资助项目(40771140/D0118)
国家重点基础研究发展规划项目(2007CB714406)资助
关键词
彩色图像
车牌检测
多尺度分割
区域生长
形状
Color image, License plate detection, Multi-scale segmentation, Region-growing, Shape
