A review on applications of deep learning in optical image cryptosystems

With the development of technology and the deep integration of digitalization and informatization, the generation and transmission of optical image data have increased exponentially. Optical image encryption, as a core technology to ensure its security, is crucial. Although traditional optical encryption has advantages such as high parallelism and multidimensional encryption, and is widely applied in fields like military reconnaissance and medical imaging, it still faces issues such as insufficient security, weak attack resistance, poor robustness, and low processing efficiency. In recent years, incorporating deep learning into optical image cryptosystems to enhance encryption strength, improve image quality, and optimize decryption results has become an important research direction in this field. This article explains the basic principles of optical image cryptosystems and the technical characteristics of each stage, analyzes the research progress and key challenges of deep learning in image encryption, quality enhancement, and decryption recovery, and finally looks forward to future development trends in aspects such as high-security and high-efficiency encryption strategies, robust decoding, cross-modal secure integration, and improving system generalization capability.