Real time character recognition of car number plates

Pattern recognition is an old and also an ongoing research area that studies the operation and design of systems that recognize patterns in data. Many powerful tools have been explored for various real-time applications such as image analysis, character recognition, speech analysis, human and machine diagnostics, person identification, industrial inspection and so on. Aiming at general objectives of real time character recognition, namely fast processing speed, high accuracy and low cost, this thesis focuses on character recognition of car number plates and aims to recognize the characters on number plate images quickly and accurately. Two different learning based methods, inductive learning and SVM based classifiers, are presented. A new combined recognition system for character recognition of car number plates is also proposed. The inductive learning based method is applied to roughly divide the candidate into primary groups corresponding to digits and letters. Then SVM based approach is used to recognize the characters in the groups accurately. Furthermore, a classification tree is proposed that provides an efficient way both in training and testing processes. Computational time is therefore saved. This approach differs from other approaches such as template matching, OCR and Neural Network based approaches. It shows efficient performance in terms of both accuracy and speed, two key indices for real time applications. Furthermore, for those number plates which were misclassified due to poor quality of number plates, a new approach of character recognition without segmentation has also been developed for quick and accurate recognition of car number plates. It achieves the perfect 100% accuracy rate on our database. In addition, based on a new image structure-Spiral Architecture (SA), modified feature masks and modified inductive learning approach are presented. A conclusion is drawn that number recognition on SA performs better than on the traditional image structure.