Number Plate Recognition without Segmentation

In real time number plate recognition, some vehicle number plates can not be recognized due to very poor illumination, motion blurred effect, fade characters and so on. The key problem is that number plate can not be segmented accurately and correctly. In this paper, we present a recognition method based on Support Vector Machines (SVMs). Firstly, some concepts of SVMs are briefly reviewed. Then a new number plate recognition algorithm is proposed. Unlike the traditional methods for number plate recognition, the innovation of the proposed algorithm is that it does not need a process for segmentation of input image of number plate but finds features in the whole number plate image. Multi-class SVMs are developed to classify the given number plate candidate. The experimental results show that our new method is of higher recognition accuracy and higher processing speed than using traditional SVM based multi-class classifier. This new approach provides a good direction for automatic number plate recognition.