Further attachment of image-based communication speed to sensitive areas must be provided with confidential and safe transmission. Conventional methods use encryption and steganography as distinct methods, hence producing either conspicuous artifacts (encryption) or attack vulnerability (steganography). Though cryptographic algorithms, including AES and DES, offer a high level of mathematical security, they generate conspicuous noise-like images, unlike steganographic algorithm schemes, including LSB and DCT embedding, which are imperceptible but have low resistance to extraction attacks. To solve this gap, we would present an all-new end-to-end convolutional neural network that fuses the image transformation and data hiding tasks into a single network. It uses the architecture of a learnable Key Mixer of adaptive, content-sensitive transformations, an Encoder of smooth embedding, and a symmetric Decoder of accurate secret recovery. There is a dual-loss operation that balances the perceptual fidelity and quality of recovery, which is effective and has the drawbacks than a conventional sequential pipeline. Thousands of experiments conducted on the STL-10 data demonstrate that the proposed model achieves PSNR values that are better than 32 dB and SSIM values that are better than 0.90, both on cover and secret images, and is therefore better than the traditional methods in both quality and fidelity to recovery. Analysis of entropy attests idealized near-randomness (~ 7.3), which is statistically insignificant. These findings show the possibility of the proposed framework as a viable and scalable architecture to deliver secure image transmission that determines a new paradigm and combines content protection with visual invisibility.
Iqbal et al. (Wed,) studied this question.