A Robust Hybrid Framework for Image Security using Chaotic Maps, Steganography, and Convolutional Neural Networks

Safeguarding sensitive visual data against unauthorized access and disclosure is critical, making robust image encryption a paramount necessity for ensuring privacy. This paper presents a novel hybrid framework for secure image coding that significantly enhances both security and privacy by integrating chaotic map-based encryption with steganography. The proposed system utilizes a sophisticated two-phase data concealment and encryption process. Initially, the sensitive image is covertly embedded within a designated cover image using the Discrete Cosine Transform (DCT) domain to achieve high imperceptibility. The resulting transformed image is then subjected to a robust encryption mechanism. This mechanism features a dynamically generated chaotic map, which is constructed by integrating multiple chaotic systems, specifically leveraging a cosine-square logistic map to produce a variable key for initiating new chaotic sequences. These highly randomized sequences are subsequently applied to encrypt the DCT coefficients of the hidden image. Furthermore, to optimize the quality and fidelity of the decrypted image, Convolutional Neural Networks (CNNs), employing diverse filter sets, are incorporated into the decryption pipeline. The efficacy of this new approach is validated through comprehensive numerical experiments conducted on standard gray-scale images. Comparative analysis against established techniques, including circular mapping, S-box implementations, and the S-box combined with the Arnold Transform, demonstrates the superior performance of our methodology. The results confirm that the suggested hybrid scheme achieves higher security metrics, including lower correlation coefficients, and excellent information entropy, significantly outperforming existing image security solutions.