Hybrid Steganography: Leveraging Chaotic Encryption and CNN for Robust Image Hiding

This paper presents a robust image information hiding method integrating Convolutional Neural Networks (CNN) and image encryption based on the Lorenz three-dimensional chaotic system. We introduce chaotic image encryption technology based on the Lorenz three-dimensional system into the CNN-based image steganography framework. Leveraging the high sensitivity and unpredictability of the Lorenz system, the original secret image is first encrypted to render it visually imperceptible. During the image hiding phase, the encrypted secret image is embedded into the carrier image using an improved CNN-based image information hiding model. This model capitalizes on CNN’s powerful automatic learning and feature extraction capabilities to achieve information hiding, resulting in a steganographic image. In the information extraction phase, the steganographic image serves as the input for the decoder network within the model, where the decoder decrypts it to retrieve both the carrier image and the encrypted secret image. Finally, the secret image is decrypted using the Lorenz chaotic encryption algorithm. Analyses of PSNR/SSIM and visual quality demonstrate that the proposed hybrid image steganography scheme offers high security and robustness, effectively preventing distortion of the secret image.