Skip to Main Content
LPR (License Plate Recognition) is a foundation component of modern transportation management systems. It uses a set of computer image-processing technologies to identify vehicle by its license plate. Character recognition is the core of LPR, which is essentially a multi-classification problem. The challenge is how to recognize every character of the license plate accurately and rapidly in case of noise, variation, blurs and other adverse conditions. As a new machine learning method, SVM (Support Vector Machine) has been used in character recognition recently and showed certain effects. To further improve the recognition accuracy and speed of LPR, we proposed a method of character recognition based on wavelet kernel LS-SVM (Least Squares SVM) in this paper. LS-SVM is an evolution of classical SVM that has higher calculating speed. Instead of the commonly used kernel function in classical SVM like RBF kernel or Gauss kernel, we use Mexico hat wavelet kernel, which has better generalizing capacity, due to the feature of multi-scale analysis and approximately orthogonal. We did experiments on nearly 500 vehicle image using our method, BP neural network method and RBF kernel LS-SVM method. Our method got a recognition rate of 98.3% and testing time of 0.13s, both were better than the other two methods.