An image steganography scheme based on ResNet

In recent years, deep learning has been used in steganography scheme, and the implemented solution had a large hidden capacity. In order to further study the effect of basic deep learning network on image steganography, a deep learning image steganography scheme based on ResNet was proposed in this paper, which applied the idea of residual blocks in ResNet to the field of image steganography. Image hiding and image extraction were realized by encoder network and decoder network. In the proposed scheme, by setting the balance parameter during training, different hiding and extraction qualities could be obtained for different scenarios. At the same time, a new performance analysis method was proposed for the field of deep steganography: the scatter plots of PSNR and SSIM. For a test set with a large amount of data, it could clearly reflect the performance of the solution, and it had a certain degree of application value in the field of deep steganography. The experimental results showed that the scheme had better invisibility when hiding, high data accuracy when extracting, and had certain advantages compared with the existing scheme. And it performed better when detected by steganalysis tools.