Lightweight Scene Text Recognition Based on Knowledge Distillation and Transformer for Various Types of ID Cards

In recent years, photo-based ID recognition technology has been rapidly adopted by the banking and finance sector, driven by advances in digitalisation and artificial intelligence. These technologies allow customers to upload their IDs via mobile apps and the banking system automatically recognises and processes the information, reducing processing times and improving customer experience. But with this convenience comes the challenge of securing sensitive ID data. Traditionally, ID data has been transmitted to external servers, which increases the potential risk of data leakage and privacy exposure. Therefore, data security is becoming more important. In this research, we propose a novel lightweight model that combines knowledge distillation and transformer technology to achieve effective performance on different types of ID cards.