Vehicle license plate character recognition based on relative entropy function criterion

Abstract

Based on gradient method,the conventional learning algorithm of Wavelet Neural Network(WNN)using the Mean Square Error(MSE)criterion may affect the convergence speed and make the process fall into local minimal.The entropy function criterion is superior to the MSE criterion,possible to improve convergence rate.Thus based on relative entropy function criterion,a novel vehicle license plate character recognition method using WNN was proposed.Firstly,image preprocessing was done on character images,including image binary,image normalization and so on.Then invariant moment of character image using the new invariant moment algorithm was extracted,which was taken as the characteristic vector.Finally,the optimized wavelet neural network was used to classify and recognize the target.Computer simulation shows that this method achieves good recognition effect.