Deepface-Based Chaotic Image Encryption Using Key Optimization and Semi-Tensor Product Theory

Currently, protecting personal privacy through selective encryption of facial images has become a research hotspot. This paper aims to design a new image encryption scheme using chaotic systems, optimization algorithms, and the Semi-Tensor Product (STP) theory. Firstly, we proposed a 3D Coupled Ikeda Map with Bounded Amplitude (3D-CIMBA), which has controllable Lyapunov exponents and high-complexity. Secondly, Particle Swarm Optimization (PSO) algorithm is used to generate control keys of the chaotic system, and produce chaotic sequences. Then, DeepFace model is applied to recognize facial regions for encryption. Moreover, the face image is encrypted by performing row-column alternation cyclic shifting operation and STP diffusion. Finally, cryptographic analysis is conducted using histograms, pixel correlation, information entropy, and SSIM. The simulation results show that this scheme demonstrates robustness against differential attacks and noise attacks, which also exhibits fast encryption speed and large key space. These results show that the proposed algorithm can encrypt facial information more efficiently and securely compared to traditional algorithms.