Ciphertext-only attack on optical cryptosystem with spatially incoherent illumination based deep-learning correlography. (March 2021)

• Record Type: Journal Article
• Authors: Wu, Huazheng; Meng, Xiangfeng; Yang, Xiulun; Li, Xianye; Wang, Pengwei; He, Wenqi; Chen, Hongyi
• Is Part Of: Optics and lasers in engineering. Volume 138(2021)
• Publication Date: 2021-03
• Subjects: Incoherent optical cryptography -- Ciphertext-only attack -- Random phase encoding -- Deep learning; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2020.106454
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: A ciphertext-only attack method relys on the statistical correlation properties of speckles, revealing that the statistical average of the Fourier transform intensity of the ciphertext sub-blocks is essentially the same as the autocorrelation of the plaintext itself is proposed. Characterize and use the ciphertext information, and use the spectrum estimation theory to derive the autocorrelation characteristics of the ciphertext information. With the process of CNN, this scheme has higher robustness and high fidelity than traditional phase retrieval algorithms. Abstract: Security analysis is crucial and indispensable to authenticate the performance of cryptographic systems. The security risks of the spatial incoherent optical cryptography system are evaluated from the perspective of scattering medium imaging, and it proves that the system is vulnerable to the pure ciphertext attacks. The proposed ciphertext-only attack method relies on the statistical correlation properties of speckles, revealing that the statistical average of the Fourier transform intensity of the ciphertext sub-blocks is essentially the same as the autocorrelation of the plaintext itself. To better model, characterize and utilize the ciphertext information, the autocorrelation is derived by using the spectral estimation theory. Then, using only the synthetic data sampled from the noise model, without knowing the plaintext, a deep convolutional neural network (CNN) is trained to conquer the noisy phase retrieval problem associated with correlography. The resulting deep-inverse correlography approach is exceptionally robust to noise, with only 4% of the ciphertext clue, the plaintext information can still be retrieved. Both the theory analysis and the experiment results validate the feasibility of the proposed ciphertext-only attack method.