Abstract
The issue of integration of optical cryptosystems and deep learning (DL) algorithms have attracted a great deal of attention recently. When neural network models pre-trained by a series of known plaintext-ciphertext pairs is utilized for decryption, however, all the physical secret keys become useless for most of reported optical encryption systems, and the security of those systems consequently relies too heavily on the network models. In this paper, we propose a sparse-data-driven framework for speckle-based optical encryption, which add an extra layer of security to DL-assisted optical cryptosystems. Two networks are trained in the encryption, one is associated with secret keys and the other is driven by sparse data. The framework we build can provide flexibility by allowing the cryptosystem to adapt the generation of cyphertext to a changeable strategy. The effectiveness and security of the proposed framework for speckle-based optical encryption is verified with different experimental results. We hope the considerations presented in this work could promote the application of DL in different types of optical cryptosystems.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call