车牌提取方法研究

提出了一种基于模糊边缘检测和纹理特征的牌照提取方法。首先由投影法确定车牌候选区域，然后由12个边缘模板计算每个候选区域中边缘两侧的平均灰度差，并作出灰度梯度的二维直方图。据此确定每个候选区域模糊边缘检测中广义渡越点的值。然后根据车牌区域特点来提取牌照。经测试，该算法简单快速、定位准确，为后继字符分割和识别做了较好的预处理工作。

On Real-Time Location of Vehicle License Plate

Real time location demands that the license plate should be located in less than 1s. We develop a method to satisfy this demand. Section 1 reviews briefly the meaning of generalized crossover point. Section 2 discusses how to select generalized crossover points for our purpose. Fig.1 in section 2 shows the twelve valid edge structures which we use to compute the absolute value of the difference of gray level averages in two regions of edge, thus obtaining the gray gradient histogram as shown in Fig.3. Section 3, based on what is presented in section 2, discusses the fuzzy edge detection of candidate regions. Fig.5 shows the result of fuzzy edge detection of the candidate regions of the car in Fig.2. Section 4 gives the details of our method for locating license plate. Location of about 400 license plates shows preliminarily that our method can indeed locate in real time a high percentage of license plates accurately-98% in good weather and 69% in poor weather.