Research Of License Plate Recognition Based On Lifting Wavelet And Support Vector Machine

Vehicle license plate recognition technology plays an important role in Intelligent Transportation System, meanwhile, it's also an integrated application of image processing, pattern recognition and computer vision technology in intelligent transportation management. In fact, the application scope of license plate recognition technology is severely constrained because of the influences of much interference in complicated conditions. Considering the excellent time-frequency feature of the lifting wavelet transform and the powerful classification ability of Support Vector Machines, vehicle license plate recognition technology based on lifting wavelet transform and Support Vector Machines is researched in the thesis after analyzing deeply the practical difficulty of every link of vehicle license plate recognition system. The main content of the thesis as follows:1. Vehicle plate location. Traditional algorithm is extremely sensitive to some factors, such as color, light, complex background, etc. So, the robustness and accuracy can not always be guaranteed. The thesis proposes a various technical fusion plate localization algorithm which is based on lifting wavelet transformation and combined with sliding model, projection algorithm, cluster analysis and mathematical morphological operation. As a result, the algorithm not only improves the locating accuracy and speed effectively, but also reduces greatly the dependence on vehicle information which is liable to vary with environment and strengthened the robustness of the algorithm.2. Character segmentation. Considering the interference factors (such as character adhesion, character fracture, etc.) and the structure feature of vehicle plate characters, vehicle plate will be divided into independent character effectively by the improved template matching algorithm and other classical algorithms including projection algorithm and connected-area algorithm, which overcomes successfully the shortcoming of single traditional segmentation algorithm which has often poor stability and low accuracy.3. Character recognition. The algorithm based on Support Vector Machine adopts the classification strategy which many features and multiple classifier are fused, meanwhile, the rough pixel character feature and the character feature based on the best basis of wavelet packet are proposed and the multi-class Support Vector Machine method based on binary tree is also improved, the experiment result shows that the algorithm become more efficient.