License Plate Character Recognition Algorithm based on Filled Function Method Training BP Neural Network

The license plate character recognition (LPCR) algorithm is considered as the most crucial step in the vehicle license plate recognition (VLPR) system. It needs fast speed in finding an optimal solution and good optimization effect. After turning it to a global optimization problem, we propose a more practicable one- parameter filled function, then present an improved LPCR algorithm which combines the filled function method and BP neural network. In the proposed LPCR algorithm, we attain a local minimizer by implementing BP neural network, and use filled function to escape the current local minimizer to a lower minimizer. Repeating these steps, a global minimizer is obtained. Some of the ideas in our method can be widely applied in pattern recognition. The application of the proposed LPCR algorithm in intelligent transportation system (ITS) in Jinhua city of Zhejiang province demonstrates faster recognition speed and greater accuracy rate compared with other methods.