An intelligent and efficient CNN-AES framework for image block encryption with a multi-key approach

Abstract The integration of cryptography and deep learning has become known as a promising way to improving image security in the context of escalating cyber threats, particularly in areas requiring secure image transmission. The proposed methodology involves a Convolutional Neural Networks model designed to encode 256 × 256 images, followed by partitioning the encoded output into 16 blocks and encrypting each block using the AES algorithm with 16 unique keys derived from an initial single key to secure image data. Extensive evaluation of the framework’s effectiveness is conducted using correlation analysis, which achieves a low correlation coefficient of approximately 0.03; high NPCR and UACI values of up to 99.4% and 51%, respectively; histogram analysis; PSNR; MSE; MAE; and the NIST test suite, among other metrics. The outcomes show that the framework is highly resistant to differential assaults and maintains minimal loss of image quality during the encryption and decryption processes. The approach addresses important issues in digital information security and unlocks the way to safer digital communications. It has major practical implications for private content sharing on social media platforms, secure medical imaging transmission, and the management of sensitive surveillance data. A comprehensive analysis shows that the proposed encryption algorithm works more effectively than the techniques presently in use for image encryption. This work highlights how deep learning and cryptography techniques can be combined to enhance image security as well as offer a robust solution to protect sensitive image data against cyber threats.