Scene text recognition & vehicle license plate recognition

• Main Author: Hu, Wen Yang
• Other Authors: Lin Zhiping
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2020
• Subjects: Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: https://hdl.handle.net/10356/140249
• Institution: Nanyang Technological University

Scene text recognition and vehicle license plate recognition both fall into the same class of computer vision problem: text recognition. Text recognition tackles the problem where characters are recognized in sequence and the length of the characters is varying. Although many recent works are proposed to improve the performance of text recognizer, there still remains a research gap on the tradeoff between the recognition accuracy and the inference speed. This project focus on the novel algorithm design on scene text recognition and the vehicle license plate recognition practice on real-world applications. Guided Training of Connectionist Temporal Classification (GTC) is proposed to achieve effective and efficient recognition. Graph Convolutional Network (GCN) is also introduced to further improve the performance. Experimental results show that my approach achieves a new state-of-the-art accuracy on most scene text recognition benchmarks and it shows great robustness on vehicle license plate recognition on real world images. A software pipeline is also made to recognize arbitrary text. Our method was even accepted as a poster in Conference AAAI 2020.