A new architecture based resnet for steganography in color images

Abstract

This paper proposes a deep-learning color image steganography scheme that employs convolutional autoencoders with ResNet architecture. In the proposed method, all images are passed through the Preproccessing model which is a convolutional deep neural network with the aim of feature extraction. Then, the Operational model generates the stego or extracted image. The advantage of the proposed structure is the identity of models in the embedding and extraction phases. The performance of the proposed method is studied using COCO and CelebA datasets. For quantitative comparisons with previous related works, the peak signal-to-noise ratio (PSNR), the structural similarity index (SSIM), and hiding capacity are evaluated. The experimental results verify that the proposed scheme performs better than traditional and previous deep steganography methods. The PSNR and SSIM are more than 35dB and 0.98, respectively which implies the high imperceptibility of the proposed method. Also, the relative capacity of the proposed method is 8 bits per pixel. To assess the robustness of the proposed scheme against various steganalysis methods, we conducted histogram, difference, and stag analyses. The outcomes demonstrated that the proposed scheme could withstand different kinds of attacks.