EFCA-DIH: Edge Features and Coordinate Attention-Based Invertible Network for Deep Image Hiding

The purpose of deep image hiding is to embed the secret image imperceptibly in an equally sized cover image, and then recover the secret image almost perfectly at the receiver end. How to improve the quality of recovered secret images while ensuring the visual quality and security of stego images is an important challenge. In order to address this issue, a novel deep image hiding framework called EFCA-DIH (Edge Features and Coordinate Attention-based Invertible Network for Deep Image Hiding) is proposed. Firstly, an important feature extraction module is proposed to extract wavelet sub-band features coupled with edge features, thereby hiding the secret image better in the cover image. Secondly, a coordinate attention mechanism is introduced into the invertible hidden module to embed the secret information in the complex texture regions. Finally, an edge feature loss function is designed to constrain the edge differences between the stego image and the cover image, and between the secret image and the recovered secret image, thereby improving the quality of both the stego image and the recovered secret image. Experimental results have demonstrated that our EFCA-DIH significantly improves the quality of recovered secret images compared with other state-of-the-art methods, while maintaining the visual quality and security of stego images.