Abstract: This paper?presents a fast, real-time?algorithm for vehicle license plate?location under complex background. The real-time and accuracy?are?the fundamental for the?license plate?recognition system. This paper proposes a special integer kernel and a difference algorithm for license plate location. Based on analyzing the hybrid Gauss function convolution method for license plate location, focusing on the higher computational complexity of the 2-dimension Gauss kernel, an improvement is made. Hybrid Gauss convolution is carried on directly after getting the vertical edge?map by sobel operation. A new integer kernel is proposed. and the convolution is replaced by the difference method in which a few addition operation is used instead of a large amount of float multiple operation. The paper compares location efficiency and time complexity for the two different methods. Experimental results illustrate that the proposed algorithm is effective for license plate location under complex background and can achieve?real-time?requirements.

Key words: license plate location, a special integer kernel, difference, Gaussian blur, mix-Gaussian function

摘要： 提出了一种复杂背景下的快速实时车牌定位算法。车牌定位的实时性和精确性是车牌识别系统高效运行的保障。提出了一种整型特殊核并利用差分运算来实现车牌定位。在分析了混合高斯函数卷积的车牌定位方法上的基础上，针对二维高斯核计算复杂度较高的问题提出了改进，直接在获取的垂直边缘图上进行混合高斯卷积，重新提出了一种整型的特征核，并用差分的方法代替了卷积，即用少量的加法操作代替了卷积大量的浮点乘法操作。对比分析了两种不同方法的定位效果和时间复杂度。实验结果表明在保持定位精度不变的情况下运行效率有较大提高，达到了实时性的要求。

关键词: 车牌定位, 整型特征核, 差分, 高斯模糊, 混合高斯函数