Forensic Reserve: Eliciting Latent Knowledge for Image Forgery Detection

2026-10-06

AI Summary

This study addresses the underutilization of latent forensic knowledge within pretrained models by proposing the RGE framework. Treating sparse internal components as a "forensic reserve," RGE employs F-lens to identify sensitive components and construct a fixed subspace. By integrating activation decomposition with subspace projection, it inserts lightweight adapters for structurally constrained training, enabling parameter-efficient fine-tuning (PEFT) without target domain adaptation. Experimental results demonstrate that, using only 500 images and fewer than 0.2% trainable parameters, the proposed method achieves competitive detection performance across three mainstream benchmarks.

Abstract

As generated images become increasingly realistic, reliable forgery detection is essential for maintaining trust in visual information. However, existing methods primarily rely on task-specific supervision to adapt vision foundation model representations, without fully exploiting internal forensic knowledge to guide detection. To address this limitation, we propose Reserve-Guided Elicitation (RGE), a framework that treats sparse, origin-sensitive internal components in pretrained models as a forensic reserve and translates their localization into structural constraints for lightweight adaptation. Specifically, we first use the Forensic Lens (F-lens) to decompose activations across layers and token groups into independent components and globally screen them by their response differences between real and generated images, identifying reserve sites and directions. Next, we map the selected directions back to hidden-state space to construct fixed reserve subspaces and insert Forensic Reserve Adapters (FRA) only at the identified sites. Finally, with the backbone parameters, previously fitted reference classifier, and subspace bases fixed, we train only the FRA coefficient maps to generate input-dependent residual updates constrained to the corresponding subspaces, strengthening existing forensic responses. Using only 500 labeled training images and a trainable parameter budget below 0.2% of the backbone, RGE achieves competitive performance across three detection benchmarks without target-benchmark adaptation. Furthermore, RGE consistently improves over the corresponding frozen detectors across eight encoders spanning self-supervised and vision-language pretraining, eliciting a latent forensic capacity broadly shared across pretrained vision models.

Problem

• Image Forgery Detection
• Vision Foundation Models
• Latent Forensic Knowledge
• Generated Images

Innovation

• Forensic Reserve
• Reserve-Guided Elicitation
• Image Forgery Detection
• Forensic Reserve Adapters
• Parameter-Efficient Adaptation