A Survey on Image Immunization Techniques

Image immunization injects imperceptible perturbations into images to raise the cost of malicious editing. Key challenges persist in balancing immunity strength and visual fidelity, achieving cross-model generalization, reducing computational overhead, and standardizing evaluation. This survey systematizes GAN-based and diffusion-based, as well as reversible-network approaches, and compiles reported results on common benchmarks to surface trade-offs among robustness, fidelity, and efficiency. We conclude with open problems and directions, including architecture-agnostic objectives, reversible carriers for authentication and near-lossless recovery, cross-modal immunity transfer, and energy-efficient substrates (e.g., quantum encodings, neuromorphic processing).