License Plate Recognition via Deep Convolutional Neural Network

Abstract

With the development of society, road vehicles have increased. Manual identification of license plates is tricky and the real-time efficiency is far less rapid than computer processing. The traditional license plate recognition technology mainly relies on the morphological processing of images. In the harsh environment, e.g. the license plate is blurred, etc.; the recognition rate is significantly reduced. In order to solve this problem of license plate recognition, we propose a new license plate recognition method. In the license plate location, we use the traditional positioning method and support vector machine (SVM) algorithm to locate the license plate. Compared with the traditional identification method - cutting and matching, we use the capability of feature extraction of convolution neural network (CNN) to identify the whole license plate directly, which could avoid the subsequent recognition error caused by the segmentation in the license plate recognition. The experimental results show that the proposed method achieves an accuracy of more than 99% in the license plate location. In the character recognition, it achieved 97.8% accuracy. The overall recognition rate is above 97%. Compared with the traditional method, our proposed method has a superior performance.