ARTIFICIAL INTELLIGENCE BASED IMAGE STEGANOGRAPHY FOR HIGH IMPERCEPTIBILITY, SECURITY AND CAPACITY

Abstract

Due to digital communication, there was an increased need of secret and secure transfer of data. In the past, it was necessary to transfer steganography of pictures to simpler methods such as LSB and DCT, whereas today, it is performed on the basis of AI-driven applications which are more noticeable, safer and empowered. In recent applications of deep learning algorithms, including CNNs and GANs, the deep learning models can be used to integrate secret information without affecting the quality and the life of the carrier. Such technologies have therefore been seen to dominate in areas like journalism and national security which should never lag behind the increasing demand of the usage. Traditional ones, however, remain vulnerable to compression, steganalysis and real-world distortions issues, which demonstrates that the development of an acceptable solutions is not a closed question so far. The paper presents a complete steganographic system, grounded on AI, a combination of adversarial training, attention models and AES encryption. To be more specific, the system has a ResNet-34 encoder, U-Net generator, PatchGAN discriminator and all three are trained on the COCO dataset. The edge detection and entropy-based region selection are the two pre-processing that are employed in the current work to get the desired outcome of the effective data embedding. These performance metrics are PSNR, SSIM, BER, BPP and detectability. The proposed model gives 42.5 dB of PSNR, 0.98 of SSIM, 0.02 of BER and 0.0156 of BPP value, which exceed LSB, DCT and DeepSteg methods. A critical part in attention modules and adversarial training was evidenced in the experiments of ablation. Its power was validated in real-world tests in locations such as IoT, blockchain and medical imaging with encodes time of less than 70 ms and higher than 95 percent recovery and low detectability (10.2) with high sensitive cases. All these findings demonstrate that the framework is a robust mechanism of steganography that is secure, hidden and high capacity. New hybrid architecture is the evolution not only in theory, but also in the side of the user, hence, the new gateways to the research of intelligent multimedia security have become open.