License Plate Detection Based on Convolutional Neural Network: Support Vector Machine (CNN-SVM)

Abstract

Automatic License Plate Recognition (ALPR) implementation can be used in many applications, such as road traffic monitoring, automatic toll payments, and parking management. License plate detection is the first and very critical stage in the ALPR system. Locating the license plate in the image becomes more difficult in the complex backgrounds such as the highways. This research develops the plate detection method in a complex environment in two stages: plate candidate extraction, and plate area selection. We use Sobel operator for vertical edge detection, closing morphological operation, and Connected Component Analysis (CCA) for contour detection in plate candidate extraction stage. Plate area selection is implemented by using Convolutional Neural Network -- Support Vector Machine (CNN - SVM). CNN acts as feature extraction method whereas SVM as a classifier. Compared to some other machine learning architecture, CNN-SVM reached the highest accuracy by 93%.