A Novel Color Image Encryption Method Based on Hierarchical Surrogate-Assisted Optimization

To address the limitations of traditional image encryption algorithms in key optimization and encryption quality assessment, in this paper we propose a framework for image encryption based on surrogate-assisted differential evolution. First, we construct a novel fitness function based on pixel correlation, which quantitatively evaluates and optimizes encryption quality by minimizing the pixel correlation coefficient. Second, we propose an adaptive hierarchical surrogate-assisted differential evolution algorithm (HSADE-IQUA), which combines global and local phases. In the global optimization phase, HSADE-IQUA significantly improves the convergence speed and solution quality in constrained optimization through adaptive parameter control. In the local optimization phase, the population size is dynamically adjusted using the exponential moving average (EMA), achieving a balance between exploration and exploitation. The performance of HSADE-IQUA has been validated on a commonly used expensive optimization benchmark suite, achieving excellent experimental results. Third, a Chen hyperchaotic-DNA coding fusion encryption framework optimized by HSADE-IQUA (HSADE-IQUA-DNA) was constructed and tested on standard computer vision images, labeled datasets, and remote sensing images, proving that HSADE-IQUA-DNA can significantly reduce pixel correlation, effectively resist exhaustive attacks, noise attacks, and shearing attacks, and accurately recover the original image. Compared with traditional chaotic image encryption, HSADE-IQUA-DNA not only has a bottleneck in parameter optimization but also alleviates the single-key issue, further improving encryption security.