Wavelet-Based Vehicle License Plate Character Recognition

Vehicle images are influenced by weather and rays, which bring geometric distortion and noise presence, etc. Then the quality of vehicle license plate characters is poorer than the hand-printed characters, which makes the traditional character recognition method not applicable. Wavelet analysis established recently is an important tool of image analysis. Its multi-scale decomposing characteristic greatly accords with human visual property, which is very similar with the cognition process from coarse to fine in computer vision. It is more fit for image information processing.This dissertation introduces the structure of the vehicle license plate recognition system firstly. The system is divided into two modules: vehicle license plate location and vehicle license plate character recognition. Then this dissertation describes several methods of the steps of vehicle license plate recognition, and discusses the problem of character isolation. This dissertation focuses on the character feature extraction methods based on wavelet. Feature extraction based on wavelet coefficients and feature extraction based on cluster of wavelet coefficients are presented. Wavelet packet decomposition is also used for feature extraction by searching the best basis. This dissertation discusses the combination of wavelet and moment function as well. Finally, gabor filters are designed for vehicle license plate character recognition, which stand out wavelet transform for self-adjusting characteristic of time-frequency window.Using BP neural network, some experimentations are implemented in this thesis. A conclusion can be drawn that the feature extraction method based wavelet packet displays the best performance.