基于CNN网络的带遮挡车牌识别

为提高存在遮挡的车牌识别准确率,基于数据驱动,利用形态学算法如腐蚀、膨胀、旋转等对标准化字符进行自动化处理,并自适应地加入高斯噪声构建带有遮挡的字符样本以代替常见的无遮挡标准车牌字符样本.结合图像边缘检测与HSV(Hue,Saturation and Value)模型对车牌实现正确定位;采取霍夫边缘检测对倾斜的车牌进行仿射校正,并归一化车牌尺寸对车牌进行规定比例的字符切分.在此基础上,使用卷积神经网络(Convolutional Neural Networks,CNN)对样本库进行训练并对车牌内容进行识别.实验结果表明,该方法对带遮挡物的车牌具有良好的识别效果,且对汉字的识别精度略高于字母及数字.通过不同网络中与无遮挡样本库的识别效果对比可知此样本库的整体识别精度确有明显提高,有一定的应用价值.

License Plate Recognition with Partial Occlusion Based on CNN

In order to improve the accuracy of license plate recognition under occlusion,morphological algorithms such as corrosion,expansion and rotation are used to automatically process the standardized characters based on data driving,and Gaussian noise is adaptively added to construct character samples with occlusion to replace common standard vehicle license plate character samples without occlusion.The image edge detection and the HSV model are combined to realize the correct positioning of the license plate.The Hough edge detection is used to affinely correct the inclined license plate,and the license plate size is normalized to the predetermined proportion of the license plate.On this basis,the convolutional neural networks (CNN) is used to train the sample database and identify the license plate content.The experimental results show that the method has a good recognition effect on the license plate with occlusion,and the recognition accuracy of Chinese characters is slightly higher than that of letters and numbers.Through the comparison of the recognition effect of the sample database in different networks with that of the unblocked sample database,it can be seen that the overall recognition accuracy of the sample database is obviously improved,which has certain application value.