基于改进 K 近邻算法的小型汽车号牌识别系统

为了对现有小型汽车号牌识别系统进行优化，改善车牌字符识别系统性能，借助 OpenCV 图像处理开源库，在车牌图像预处理阶段采用均值滤波方法提高图像质量，采用 Sobel 边缘检测算子对图像边缘进行提取，利用交替的膨胀、腐蚀操作结合车牌长宽比实现车牌轮廓定位，并根据列像素值对车牌字符进行切割，最后采用改进的 K 近邻算法对分割后的单个车牌字符进行识别。实验结果表明，基于改进 K 近邻算法的车牌识别系统处理时间为 2.08s，识别正确率达 91.3%。与传统的 K 近邻算法相比有着更高的识别率，与神经网络法相比，有着更快的识别速度。

License Plate Recognition System Based on Improved K Nearest Neighbor Algorithm

In order to optimize the existing license plate recognition system，improve the performance of the license plate character recognition system，in this paper，with the aid of OpenCV image processing open source library，in the license plate image preprocessing stage，we use the average filtering method to improve the images’quality，adopt Sobel edge detection operator to extract images’ edge，achieve the license plate contour positioning through the use of alternating dilation and erosion，and cut license plate character according to the column pixels. At last，the improved K-nearest neighbour algorithm is used to recognize the single plate character after segmentation. Experimental results show that the processing time of the vehicle license plate recognition system based on the improved K-nearest neighbor algorithm is 2.08 seconds，and the recognition accuracy is 91.3% . Compared with traditional K-nearest neighbor algorithm，this algorithm has higher recognition rate，and it has faster recognition speed than neural network.