New data hiding algorithm using generative adversarial networks to enhance security level and payload capacity

Steganography hides essential and secret information, like images or videos, within a cover medium. It has long been the cornerstone of secure communication. With the advent of digital media, steganography technology has expanded to include various techniques and applications, from digital watermarking to multimedia. Generative Adversarial Networks (GAN) have recently appeared as a powerful technique for steganography, enabling the creation of realistic, imperceptible cover media that can effectively hide confidential information. Despite significant progress and extensive research, challenges must be addressed in achieving high extraction accuracy, efficient training, and diverse steganographic images. This paper aims to enhance steganography accuracy and efficiency using Dynamic Selective Layer Training and Custom Progress Growing Manager Based on Loss Function Rating in Dynamic Selective ProGAN (DSPGAN). This improves the Generative Adversarial Networks (GAN) training efficiency and stability. The proposed approach is to implement a layer-by-layer training technique to increase the diversity and quality of generated images. Experimental results show that our improved Generative Adversarial Networks (GAN) model achieves high accuracy in embedding and extraction, reduces training time, and produces non-contrasting and diverse images from stealth. This study provides a robust framework for efficient and secure stealth, leveraging the strengths of GAN while mitigating their weaknesses.