Image Steganography with Security Using Massive Threefold Attentional Residual GAN Optimized By Chaotic PSO Algorithm

The purpose of this study is to address the persistent challenges in image steganography, namely suboptimal feature learning, mode collapse, and training instability, which limit the performance of existing CNN- and GAN-based approaches for secure communication. To overcome these issues, a novel framework called Massive Threefold Attentional Residual GAN (MTARGAN) is proposed, in which the GAN hyperparameters are dynamically optimized using a Chaotic Particle Swarm Optimization (CPSO) algorithm. This design enhances feature extraction, embedding efficiency, and robustness against steganalysis. Experimental evaluations demonstrate that the proposed model achieves superior imperceptibility and resilience compared to state-of-the-art methods, with average PSNR values of 36.06, 34.43, 30.05, and 33.92 dB and corresponding SSIM scores of 0.96, 0.86, 0.89, and 0.84 at embedding capacities of 1, 2, 3, and 4 bpp, respectively. These results highlight the model’s ability to maintain a balance between embedding capacity and image quality while ensuring high recovery accuracy and security. Overall, the findings suggest that MTARGAN with CPSO optimization offers a stable, robust, and secure solution for practical image steganography applications.