Deep Learning Iris Cryptography: A Deep Learning-Based Image Encryption Framework for Secure Information Management

As high-resolution image data increasingly becomes commonplace in healthcare, surveillance systems, smart environments, and cloud platforms, ensuring image security has become one of the major challenges of our time. In this paper, we develop DL-IrisCrypt, a novel biometric-based image encryption framework in which the ciphertext is associated with the identity of the authorized user. We extract a ۲۰۴۸-bit iris code using Daugman segmentation and Gabor filters, which is then used to generate a key for initializing a six-dimensional hyperchaotic system and processed through SHA-۵۱۲ and HKDF. To enhance randomness in the key-generation process, we incorporate features extracted from the latent space of a properly trained U-Net autoencoder model. The encryption architecture uses Fisher-Yates permutation, two layers operating with PCG۶۴ in counter mode, and three passes of BLAKE۲b cipher-block chaining. Experimental results show that the proposed approach achieves an NPCR of ۹۹.۶۱%, a UACI of ۳۳.۴۶%, and an information entropy of ۷.۹۹۶۶ bits/byte, while passing all randomness tests provided by NIST SP ۸۰۰-۲۲. The framework requires an average of ۱۶ milliseconds for encryption and ۱۸۸ milliseconds for end-to-end processing on a single GPU and has an effective key space of ۲^۵۱۲. In addition to providing robust information security, DL-IrisCrypt offers features that can benefit organizations handling sensitive visual materials. By combining biometric identification, deep learning, and lightweight encryption mechanisms, the system enables secure management and preservation of the confidentiality of important digital assets while facilitating efficient information sharing with low computational requirements.

کلیدواژه ها:

Deep learning ، image encryption ، iris biometrics ، SHA-۵۱۲ ، U-Net autoencoder ، BLAKE۲b-CBC