A New Convolutional Architecture for Vietnamese Car Plate Recognition

Abstract

License plate recognition is a form of an intelligent transportation system. Although, there have been many studies on plate detection, character segmentation, and character recognition, many challenges have still remained. Convolutional Neural Network (CNN) has proven to be a powerful classification tool to achieve state-of-the-art results on various recognition tasks. In the problem of number plate recognition, CNN based methods are being used to solve problems such as plate detection, character segmentation, and character recognition. Quality of identification depends on the quality of each task. Viet Nam does not have a recognition system that combines the three tasks together. So, our key idea is to combine detection, segmentation, and recognition of multi-character number plates using CNN. Our purpose is to recognize the full sequence of the number plate without pre-segmentation. This paper presents a CNN-based method for high accuracy car license plate recognition. The presented methods are evaluated 1,000 plate images of US car plates and 1,000 plate images for Vietnamese car plate recognition. The experimental results show that our network achieves better performance than many standard plate detection and recognition algorithms. This dataset and the investigation results could be used as a baseline for future research in the field.