Indonesian License Plate Recognition Using Convolutional Neural Network

Abstract

License plate is a part of vehicle's identity. In modern countries, license plate recognition has been developed to collect traffic activity information. The performance of license plate recognition system tend to drop when the input picture contains noises like illumination, dirt, and scratches which cover one or more characters in the license plate. This research was focused on Indonesian license plate as many license plates in Indonesia had various noises like plastic cover and scratches which complicate the recognition. In this study, Indonesian license plate recognition is formed using a Convolutional Neural Network (CNN) which is known to have good performance in recognizing objects, even though the objects are obscured to some degree. Sliding window is used in this study for replace character segmentation. CNN will predict images in every area of window. The highest performance for the whole system to the normal data test is 87.36% and noised data test is 44.93%.