A Transfer learning model for high secure image steganography using ResNet-50

Abstract

Cryptography is commonly employed in data security, but one of the main disadvantages of the scheme would be that it attracts attention to just the perpetrator, which results throughout the exposure of the data. In order to solve these issues, digital image steganography is among the several techniques used to provide protection during data transmission. Capacity embedding and distortion are indeed the two key factors that play a key role when hiding a message in image. Therefore, this paper explores strategies consisting of two steps leading to the creation of that very stable framework with a large embedding ability. The first step determines an arbitrary image by morphing through hybrid discrete wavelet transform (HDWT) in order to produce a cover image that relies on the characteristics of both the chosen image. That second step integrates a text message of adequate duration using the transfer learning based ResNet50 architecture, integrating multi-keys and compound operators into the algorithm to provide robust protection. Simulation results shows that the proposed method resulted in superior PSNR and SSIM performance as compared to state of art approaches.Finally, the robustness and imperceptibility of the proposed method is improved, which reduced the attacks on the system.