Security Analysis of DeepKeyGen Based Medical Image Cryptosystem

This research paper provides a comprehensive security analysis of a newly proposed deep learning-based generator for medical image encryption and decryption stream ciphers. The proposed scheme involves the use of a DeepKeyGen network, which utilizes a Generative Adversarial Network (GAN) as its learning framework to generate private keys. Additionally, the network is guided to learn the mapping from initial images to private keys by designing a transformation domain. This ensures that the necessary keys are generated for the encryption and decryption of the medical images. However, the study found that the scheme has some security vulnerabilities due to low sensitivity in key design and generation. These vulnerabilities were confirmed through simulation experiments and analysis. In conclusion, the paper provides constructive suggestions for improving the scheme and offers guidance for future deep learning image encryption research.