A High-capacity Image Hiding Method Based on U-shaped Swin Transformer and Wavelet Transform
Wenxuan Li, Ying Liu, Tingge Zhu · 2023
To solve the issues of low hiding capacity and poor visual quality in current image hiding methods, this paper proposes a high-capacity image hiding method based on U-shaped Swin Transformer and wavelet transform. The proposed method takes advantage of the network structure features of UNet and the benefits of Swin Transformer in image processing, constructing an SAUNet network as the main structure for hiding and extraction networks. This paper proposes an MCAM attention mechanism incorporated into skip connections to further enhance image quality. The proposed method first hides the large secret image wavelet coefficients in the small cover image through a hiding network, resulting in a hidden image. In the extraction network, the SAUNet network extracts the wavelet coefficients of the secret image from the hidden image and then restores the secret image by inverse wavelet transform. Experimental data indicate that the proposed method not only enhances the hiding capacity, but also keeps high image quality. Compared to methods with the same hiding capacity, the proposed method exhibits significant improvements in visual appearance and objective evaluation metrics, such as PSNR and SSIM.