Recognition of Unsegmented and Challenging License Plate Based on Multi-scale Information Convolutional Neural Network

To automatically recognize blurred license plate images produced in extreme weather conditions, a new recognition network structure based on the indivisible highway challenging license plate was proposed. The network structure consists of feature extraction network and recognition framework. Experimental research was carried out with the Southeast University Blurred Chinese License Plate Dataset (SBCLPD). And the results showed that the Inception V3 features combined with CNN recognition framework offered better performance than other five counterparts, which achieved over 89.62% accuracy in comparative experiments.