A Region-Selective Anti-compression Image Encryption Algorithm Based on Deep Networks

Abstract

Abstract In recent years, related research has focused on how to safely transfer and protect the privacy of images in social network services while providing easy access by authorized users. To safeguard privacy, we suggest an image encryption scheme that combines data hiding and image encryption. The proposed scheme successfully decrypts images after JPEG compression attacks and preserves the privacy of secret regions through the use of block scrambling encryption based on region selection. Simultaneously, the scheme can handle nonuniform secret regions and obtain more sensitive secret keys because of the incorporation of a chaotic system. The enhanced deep learning-based data-hiding technology reduces algorithm complexity by enabling the encryption position to be determined in the decryption phase without the need for any information or equipment. However, this approach also increases algorithm security, because only when the right secret data are extracted can they be decrypted successfully. According to the experimental findings, the proposed scheme can correctly decrypt images via JPEG compression while maintaining visually acceptable quality. The proposed scheme can achieve greater robustness against image processing algorithms and a wider secret key space than traditional schemes.

Keywords

• JPEG compression
• Region selection
• Data hiding
• Block scrambling
• Deep learning