Image steganography is a scheme that hides secret information in a cover image without being perceived. Most of the existing steganography methods are more concerned about the visual similarity between the stego image and the cover image, and they ignore the recovery accuracy of secret information. In this paper, the steganography method based on invertible neural networks is proposed, which can generate stego images with high invisibility and security and can achieve lossless recovery for secret information. In addition, this paper introduces a mapping module that can compress information actually embedded to improve the quality of the stego image and its antidetection ability. In order to restore message and prevent loss, the secret information is converted into a binary sequence and then embedded in the cover image through the forward operation of the invertible neural networks. This information will then be recovered from the stego image through the inverse operation of the invertible neural networks. Experimental results show that the proposed method in this paper has achieved competitive results in the visual quality and safety of stego images and achieved 100% accuracy in information extraction.

With the development of digital imaging devices, the process of recording sensitive information displayed on screens through mobile phones and cameras has become a prominent technique for modern data leaks. In order to identify the origin of information violations, Screen-Shooting Resistant Watermarking (SSRW) has attracted a lot of attention. Most existing solutions are based on Convolutional Neural Networks (CNNs) for the embedding of watermarks. However, due to the limited reception field of CNNs, they are proficient in extracting local features but cannot understand the entire image. This paper presents a new watermarking system that is resistant to screen recording, with multi-head and cross-attention to incorporate watermarks, replacing the encoder in the end-to-end architecture. Specifically, we segment the image and watermark into smaller patches for positional embedding. Afterward, we calculate the attention scores through multi-head attention layers and generate the encoded image through concatenation. This approach increases the model’s ability to comprehend the entire image, thereby increasing performance. In addition, we enhance the U-Net network structure to replace the end-to-end decoder. The experimental results demonstrate that the proposed method not only reaches more than 95% accuracy in different capture scenarios but also excels in terms of reliability and invisibility relative to current state-of-the-art (SOTA) methods. In addition, this approach yields impressive PSNR and SSIM average values of 41.90 dB and 0.99, showing the excellent visual quality and reliability of the watermarked images.