基于 Gabor 滤波多特征融合的车牌定位算法-《计算机技术与发展》

文章信息/Info

Title:: A Vehicle License Plate Localization Algorithm Based on Gabor Filtering and Multi-feature Fusion

Author(s):: YU Hai-he; YANG Shuo; LI Da-zhou; School of Computer Science and Technology,Shenyang University of Chemical Technology, Shenyang 110020,China

Keywords:: Gabor filtering; license plate area detection; multi - feature fusion; feature point detection; gray - level projection; vehicle license plate location

摘要:: 车牌定位是车辆监控的基础。单一特征在特定的场合中定位准确,但是在实际生活场景中单一特征存在着很多缺陷。基于上述问题,提出了一种基于 Gabor 滤波器多种特征融合的车牌定位算法。该算法在实现过程中主要分为三个阶段,第一阶段是根据车牌图像大小和特点生成 Gabor 滤波核,使用 Gabor 滤波器对车牌图像进行滤波,生成车牌图像的特征图,再对滤波后的图像使用数学形态学操作,将车牌区域连接形成连通区域。第二阶段通过车牌的长宽比以及车牌字符个数进行筛选,得到车牌的候选区域,然后再使用车牌的字符纹理和字符角点邻域颜色,在候选区中进行筛选,生成候选区域。第三阶段验证候选,使用颜色方差和灰度投影确定候选。实验表明,在该图像库中,车牌区域定位的成功率达到了 95.7% ,证明了该算法是一种可行性较高的车牌定位算法。

Abstract:: License plate location is the basis of vehicle monitoring. Single feature is located accurately in specific occasions,but there are many defects in real life scenes. Based on the above problems, we propose a license plate location algorithm based on Gabor filter multifeature fusion. The algorithm is mainly divided into three stages in the implementation process. The first stage is to generate Gabor filter kernel according to the size and characteristics of license plate images. Gabor filter is used to filter the license plate image to generate the feature map of the license plate image,and then mathematical morphology is used to connect the license plate regions to form connected regions. In the second stage,the candidate regions of the license plate are obtained by screening the length-width ratio of the license plate and the number of license plate characters,and then the candidate regions are generated by screening in the candidate regions using the character texture and the color of the character corner neighborhood of the license plate. In the third stage,the candidates are verified. The color variance and gray projection are used to determine the candidate. Experiment shows that the success rate of license plate location in the image library reaches 95.7% ,which proves that it is a feasible license plate location algorithm.