Color Image Steganography Based on Two-Channel Preprocessing and U-Net Network

The advancement of technology provides convenience for information transmission, but there are still security issues in the information transmission process where secret information may be stolen. With the increasing application of the U-Net network, a novel image steganography technique based on the U-Net structure is designed. The special two-channel preprocessing network is designed to fuse image features, and the SENet attention mechanism has been introduced to the model. To better optimize the model, a mixed loss function is constructed that automatically adjusts the weights between mean square error and structural similarity. During the information transmission process, the sender can use the hiding network to conceal the secret image in the cover image and then send the hidden image to the recipient. When receiving the hidden image, the recipient can employ the extracting network to recover the secret image. The experimental outcomes indicate that the proposed model increases the visual effect of images, with cover images PSNR and SSIM reaching 40.36 dB and 98.18%, respectively. Therefore, the model can effectively hide images during information transmission and prevent attackers from discovering them.