Improved Automatic License Plate Recognition in Jordan Based on Ceiling Analysis

Abstract

The main challenge of automatic license plate recognition (ALPR) systems is that the overall performance is highly dependent upon the results of each component in the system’s pipeline. This paper proposes an improved ALPR system for the Jordanian license plates. Ceiling analysis is carried out to identify potential enhancements in each processing stage of a previously reported ALPR system. Based on the obtained ceiling analysis results, several enhancements are then suggested to improve the overall performance of the system under study. These improvements are (i) vertical-edge histogram analysis and size estimation of the candidate regions in the detection stage and (ii) de-rotation of the misaligned license plate images in the segmentation unit. These enhancements have resulted in significant improvements in the overall system performance despite a <1% increase in the execution time. The performance of the developed ALPR is assessed experimentally using a dataset of 500 images for parked and moving vehicles. The obtained results are found to be superior to those reported in equivalent systems, with a plate detection accuracy of 94.4%, character segmentation accuracy of 91.9%, and character recognition accuracy of 91.5%.

Keywords

• character recognition
• character segmentation
• intelligent transport system
• license plate recognition
• machine learning
• neural networks