Anti-tampering method of QR code hidden information based on deep learning

QR codes have gained widespread adoption across various industries for their large data capacity, rapid scanning speed, and robust error correction capabilities. However, they also face security risks, such as being copied or forged, which can lead to information leakage. To address these challenges, this paper proposes a QR code information hiding and tamper-proof mechanism that integrates convolutional neural network and twin networks. This mechanism designs and implements a robust and invisible information watermark embedding method using deep learning algorithms. It also constructs a similarity evaluation model based on twin networks for watermark extraction, accurately judging the authenticity of QR codes and effectively resisting forgery and tampering. In the analysis of the experimental results, the normalization factor between the original QR code image and the QR code image with embedded secret information reached over 0.99, and the Peak Signal-to-Noise Ratio exceeded 42 dB, significantly demonstrating the invisibility of the secret information. In robustness evaluations, the QR code can withstand Gaussian noise and image pollution attacks up to 30% without affecting the extraction of secret images. In the similarity assessment of secret images, the normalization coefficient and Peak Signal-to-Noise Ratio exceed 0.99 and 37 dB, respectively. Furthermore, the similarity between the original secret image and the extracted secret image, as detected by the twin network, exceeds 0.99, fully highlighting the superiority of this scheme. This approach is both effective and practical in ensuring the information security of QR codes, offering a novel solution for their secure application. The code is available at the following https://github.com/JCandyGold/CNN_QR_watermarks. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.