Visually meaningful image encryption using convolution neural networks

Image encryption techniques are broadly used to protect confidential images during transmission and storage. However, conventional image encryption techniques typically produce noisy ciphertext images that can easily reveal the presence of encrypted information and attract potential attacks. Visually meaningful image encryption (VMIE) technique has emerged to address this problem as an effective solution by embedding encrypted data into a cover image. Hiding encrypted data within a cover image provides an additional layer of security compared to traditional encryption methods. This property makes VMIE suitable for covert communication scenarios where both data confidentiality and transmission imperceptibility are required. In this work, a VMIE technique utilizing Convolution Neural Network (CNN) is proposed. In the first phase, a chaotic image encryption scheme refined using a CNN is used to generate highly secure encrypted images with enhanced randomness. The CNN-based refinement module learns nonlinear transformations of the combined chaotic sequences and enhances their randomness characteristics. CNN helps to remove residual correlations and improves the statistical properties of the generated random sequence by capturing complex nonlinear dependencies within the chaotic values that contribute to stronger security performance of the encrypted images. In the next phase, an output image from first phase is entrenched into a cover image through CNN based encoder to generate visually meaningful ciphertext (VMC). The proposed method achieves an overall key space greater than 2200, that ensures strong resistance against brute-force attacks. The experimental results demonstrate high differential attack resistance with NPCR values reach up to 99.61%. The VMCs maintain excellent perceptual quality up to 43.8 dB PSNR and SSIM values above 0.99 with relative payload capacity of 1. These results validate the robustness, imperceptibility, and practical applicability of the proposed VMIE scheme.