SafeCoder: A machine-learning-based encoding system to embed safety identification information into QR codes

Abstract

In social networks, the Internet of Things, mobile computing, electronic commerce, and other fields, Quick Response (QR) codes have been widely used as the interface between online and offline scenarios. In offline lives, users can readily scan QR codes with smartphones to access online networks or remote devices. However, standard QR codes appear as random black-and-white modules that make users difficult to visually distinguish the encoding message in QR codes, which incurs users unwittingly to scan malicious QR codes distributed by hackers, and results in serious issues of cyber security and privacy leakage. In this paper, we propose a novel system that can embed safety identification information (SII) into QR codes, which conduces to avoid issues of cyber security and privacy leakage. Although some existing methods attempt to embed visual information into QR codes, these methods require complex pre-designed algorithms and leave some limitations in visual representation. Unlike them, our system employs the machine learning technique, which can naturally embed SII into QR codes without compromising the scanning-robustness and message preservation. The decoding time of our results is an average of 0.72 s which is similar to the standard QR code. Subjective and objective experiments show that our system can effectively produce embedded QR codes that are applicable in real-world life, and reach the state-of-the-art level.