Decryption of Deterministic Phase-Encoded Digital Holography Using Convolutional Neural Networks

Abstract

Digital holographic encryption is an important information security technology. Traditional encryption techniques require the use of keys to encrypt information. If the key is lost, it is difficult to recover information, so new technologies that allow legitimate authorized users to access information are necessary. This study encrypts fingerprints and other data using a deterministic phase-encoded encryption system that uses digital holography (DPDH) and determines whether decryption is possible using a convolutional neural network (CNN) using the U-net model. The U-net is trained using a series of ciphertext-plaintext pairs. The results show that the U-net model decrypts and reconstructs images and that the proposed CNN defeats the encryption system. The corresponding plaintext (fingerprint) is retrieved from the ciphertext without using the key so that the proposed method performs well in terms of decryption. The proposed scheme simplifies the decryption process and can be used for information security risk assessment.