Intelligent plate number recognition system using segmentized method with artificial neural networks

This thesis research is concerned with the field of recognition of an image that used an automated plate number recognition system (APNR). In this thesis, the proposed system for the recognition of vehicle license plate numbers is designed with machine learning algorithms. The APNR system is an image processing technology that uses the license plate number to recognize/identify a vehicle. One of the objectives is to design an efficient, effective automatic vehicle identification system by using vehicle license plate number. One of the techniques that can be used to achieve this goal is the optical character recognition (OCR) system used for character recognition. The system was designed and simulated in MATLAB environment and its performance was tested on real plate images. It was observed that the system developed in experimental studies effectively detects and recognizes license plate images. The system captured the vehicle license plate images and send them to the computer where they are processed and identified. Different algorithms such as localization, segmentation, and OCR are used in the processing of the captured images. Artificial neural networks are a powerful tool that can capture and characterize complex input / output relationships. The reason behind the development and use of artificial neural networks technology stems from the need to develop an artificial system that can perform intelligent tasks similar to those performed by the human brain. Feed-forward back propagation networks are created by generalizing the gradient descent learning rule to multilayer networks and nonlinear differentiable activation functions. Back propagation is a method for calculating the gradient for nonlinear multilayer networks. This method has many variations based on other standard optimization techniques such as Newton and conjugate gradient methods. Properly trained back propagation networks tend to give reasonable responses to input they have never seen. The variations of these back propagation networks are used for character recognition and comparative results on the performance of different algorithms are presented in this thesis.