Double random phase encryption (DRPE) system is a simple and powerful encoding technique that consists of only two lenses and two random phase masks. However, there are many issues for applying to actual security systems such as phase acquisition, vulnerability to phase retrieval techniques, and data throughput. Although various extensions of DRPE have addressed each issue, there is no comprehensive solution. To tackle all the issues of DRPE, we propose a new amplitude-based DRPE (ADRPE) system using deep learning. The encoding is the same as the current ADRPE system, and the decoding is achieved by an inverse ADRPE system using convolution neural networks. Our system can achieve a real-time end-to-end encryption system without any additional optical devices and exposure of the keys. To demonstrate our method, we applied it to simulations with various datasets such as passwords, Quick-Response (QR) codes, and fingerprints.

双随机相位加密 (DRPE) 系统是一种简单而强大的编码技术，仅由两个镜头和两个随机相位掩码组成。然而，应用于实际安全系统存在许多问题，例如相位获取、相位检索技术的脆弱性和数据吞吐量。尽管 DRPE 的各种扩展已经解决了每个问题，但没有全面的解决方案。为了解决 DRPE 的所有问题，我们提出了一种使用深度学习的新的基于幅度的 DRPE (ADRPE) 系统。编码与目前的ADRPE系统相同，解码是通过使用卷积神经网络的逆ADRPE系统实现的。我们的系统可以实现实时端到端加密系统，无需任何额外的光学设备和密钥的暴露。为了演示我们的方法，