Abstract: Under complex environment it is difficult to achieve high performance of vehicle license plate localization. A template probability model-based approach is proposed. There are two stages in the proposed method: candidate generation and probabilistic verification. In the first stage, spatial distribution of detected feature points is combined with color information to generate possible candidate region of vehicle license plate; in the second stage, the structure and geometric of standard vehicle license are employed to construct similarity measure probability function. Then the localization of vehicle license plate is detected by extracting maximum similarity measure probability from candidate region. Experimental result shows high performance of the proposed method, such as license plate successful detection rate beyond 96.2 % and miss rate less 3.8%. A novel Chinese license plate localization algorithm was proposed here, which shows low complexity, but high accuracy and practicability. So that it can be applied to license plate recognition.

Key words: License plate localization, probabilistic model template, multiple features fusion

摘要： 针对现有理论解决复杂环境下车牌图像准确定位效果不佳的问题，提出了一种基于模板概率密度函数的车牌定位方法。该方法将定位过程分为车牌候选区域检测和最大相似概率定位两个阶段：在候选区域检测阶段，利用特征点的空间分布情况与颜色信息快速地确定车牌候选区域；在准确定位阶段，根据我国标准车牌的结构特点与几何信息，构造出标准车牌相似性概率密度函数。通过计算和比较候补区域中每个位置的车牌相似性概率值来实现准确定位。实验表明，提出的算法定位精度高（96.2 %），鲁棒性强（漏检3.8 %），并且实现简单。能够快速而准确地完成国内车牌的准确定位，在车牌识别领域中具有很好的实用价值。

关键词: 车牌定位, 模板概率密度模型, 多特征融合