Nonlinear bit-level image encryption method integrating DCGAN-derived dual-branch features and chaotic regulation

Abstract To address the core bottlenecks of insufficient nonlinear adaptive capability of chaotic systems and the difficulty in balancing security and computational efficiency in bit-level transformation for existing image encryption methods, this paper proposes a nonlinear bit-level image encryption scheme combining a dual-branch feature regulation mechanism based on Deep Convolutional Generative Adversarial Networks (DCGAN) and an adaptive chaotic system. First, a key-controlled dual-branch DCGAN fusion module is constructed to generate cipher keys tightly correlated with plaintext content adaptively. Second, an adaptive Chen chaotic system modulated by plaintext features is designed; comprehensive dynamical analysis verifies its outstanding nonlinearity and randomness. Third, a low-complexity nonlinear bit-level transformation is elaborated to simultaneously achieve powerful confusion-diffusion capability and low computation cost. Experimental results demonstrate that the key space of the proposed method reaches 2 256 , and its security performance outperforms mainstream comparative algorithms. Meanwhile, the computational efficiency is improved by 29.7% compared with traditional DNA-based bit-level schemes, which provides a new perspective for the collaborative encryption of nonlinear dynamics and deep learning.