基于边缘颜色对特征及笔画穿越双层检测车牌定位算法

针对现有车牌定位算法鲁棒性不够、准确度不高以及参数设置困难等问题,提出基于边缘颜色对特征以及笔画穿越双层检测车牌定位算法,不但充分利用车牌边缘颜色搭配信息,而且有效利用了车牌字符结构信息。粗检测阶段：首先进行边缘检测,人工收集所有搭配的彩色边缘特征数据,利用机器学习模型建立车牌边缘颜色对覆盖分类学习模型,然后利用车牌边缘颜色对覆盖分类学习模型,并利用先验信息进行形态学处理形成车牌候选区域。验证阶段：针对粗检测车牌候选区域,扫描车牌边缘穿越信息,最后利用车牌区域整体边缘分布覆盖分类模型进行候选区域验证处理。该方法利用车牌背景与字符具有固定颜色搭配的重要特点,综合利用了车牌的结构特征和纹理特征,提高了车牌定位的可靠性。实验采用100幅含有不同颜色搭配的车牌图像进行实验,定位准确率达到96%以上。

A vehicle license plate locating algorithm based on edge-color pair feature and stroke traversing double detection

Aiming at the robustness and accuracy of the existing license plate localization algorithm is not high enough and parameter setting is difficult, a vehicle license plate locating algorithm based on edge-color pair feature and stroke traversing double detection algorithm is presented in this paper. The algorithm not only makes full use of license plate edge color collocation feature but also uses character struc~re feature at the same time. In preliminary positioning module, the edge is detected firstly, character data of all the color edge collocation is collected artificially. Edge-color pair feature coverage model is constructed through machine learning. Secondly, the constructed model is used and the plate candidate area is shaped after morphology processing by using prior information. In verification module, edge traversing information is scanned in the license plate candidate area. Finally, the license plate candidate area is validated using the license plate overall regional edge distribution coverage model. The proposed method fo- cuses on matching background color and character color in a license plate and combines its structure feature and texture feature to improves the reliability of the license plate location. One hundred images of containing different color collocation are tested and positioning accuracy is more than ninety-six percent.