Date of Award
Electrical and Computer Engineering
Optical Character Recognition system (OCR) has been found very useful in the field of intelligent transportation. In this work, a FPGA-based OCR system aimed at image-based License Plate Recognition (LPR) has been designed and tested. A feed-forward neural networks has been chosen as the core of the proposed OCR system. The neural network parameters are acquired beforehand and will not change during its operation time. A set of Matlab programs have been made in the network's design process. The verification process includes Matlab simulation where programs using binary numbers which has the same representation format as the system are used to compute the results, and Modelsim simulation where data is computed and transferred between modules under clock signals' control. The synthesis process is done in the Altera's FPGA design software - Quartus II. The result shows that calculation speed of the system implemented in hardware is much faster than software running on a PC while it maintains a high recognition accuracy. The proposed image recognition system is used with a set of images that are generally difficult for such networks to handle. Most images include shadows and other imperfections in the image. The proposed network was able to achieve $95 \%$ accuracy in recognizing the correct character despite the image imperfections. Moreover, it takes advantage of very compact and efficient non-liner sigmoid function.
Jing, Yuan, "An Efficient FPGA Implementation of Optical Character Recognition System for License Plate Recognition" (2016). Electronic Theses and Dissertations. 5832.