Template neural particle optimization for vehicle license plate recognition

A Jalil, Norazira (2015) Template neural particle optimization for vehicle license plate recognition. Masters thesis, Universiti Teknikal Malaysia Melaka.

[img] Text (24 Pages)
Template Neural Particle Optimization For Vehicle License Plate Recognition.pdf - Submitted Version

Download (271kB)
[img] Text (Full Text)
Template neural particle optimization for vehicle license plate recognition.pdf - Submitted Version
Restricted to Registered users only

Download (2MB)


The need for vehicle recognition has emerged from cases such as security, smart toll collections and traffic monitoring systems. This type of applications produces high demands especially on the accuracy of license plate recognition (LPR). The challenge of LPR is to select the best method for recognizing characters. Since the importance of LPR arises over times, there is a need to find the best alternative to overcome the problem. The detection and extraction of license plate is conventionally based on image processing methods. The image processing method in license plate recognition generally comprises of five stages including pre-processing, morphological operation, feature extraction, segmentation and character recognition. Pre-processing is an initial step in image processing to improve image quality for more suitability in visualizing perception or computational processing while filtering is required to solve contrast enhancement, noise suppression, blurry issue and data reduction. Feature extraction is applied to locate accurately the license plate position and segmentation is used to find and segment the isolated characters on the plates, without losing features of the characters. Finally, character recognition determines each character, identity and displays it into machine readable form. This study introduces five methods of character recognition namely template matching (TM), back-propagation neural network (BPNN), Particle Swarm Optimization neural network (PSONN), hybrid of TM with BPNN (TM-BPNN) and hybrid of TM with PSONN (TM-PSONN). PSONN is proposed as an alternative to train feed-forward neural network, while TM-BPNN and TM-PSONN are proposed to produce a better recognition result. The performance evaluation is carried out based on mean squared error, processing time, number of training iteration, correlation value and percentage of accuracy. The performance of the selected methods was analyzed by making use real images of 300 vehicles. The hybrid of TM-BPNN gives the highest recognition result with 94% accuracy, followed by the hybrid of TM-PSONN with 91.3%, TM with 77.3%, BPNN with 61.7% and lastly PSONN with 37.7%.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Neural networks (Computer science), Image processing, Neural Particle Optimization, Vehicle License Plate Recognition
Subjects: Q Science > Q Science (General)
Q Science > QA Mathematics
Q Science > QA Mathematics > QA76 Computer software
Divisions: Library > Tesis > FTMK
Depositing User: Mohd Hannif Jamaludin
Date Deposited: 02 Aug 2016 06:19
Last Modified: 01 Jun 2022 15:30
URI: http://eprints.utem.edu.my/id/eprint/16822
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item