Search papers, labs, and topics across Lattice.
This paper introduces a novel "courtroom-style adjudication framework" for image manipulation localization (IML) that explicitly models evidence for both manipulated and authentic regions. The framework uses a dual-hypothesis segmentation architecture with prosecution and defense streams, guided by edge priors and disagreement suppression, to generate evidence. A reinforcement learning judge model then strategically refines uncertain regions based on advantage-based rewards and soft-IoU, improving localization accuracy.
By framing image manipulation localization as a courtroom trial between manipulated and authentic evidence, this work achieves state-of-the-art performance, even when manipulation traces are subtle.
Although some existing image manipulation localization (IML) methods incorporate authenticity-related supervision, this information is typically utilized merely as an auxiliary training signal to enhance the model's sensitivity to manipulation artifacts, rather than being explicitly modeled as localization evidence opposing the manipulated regions. Consequently, when manipulation traces are subtle or degraded by post-processing and noise, these methods struggle to explicitly compare manipulated and authentic evidence, resulting in unreliable predictions in ambiguous areas. To address these issues, we propose a courtroom-style adjudication framework that regards IML task as the confrontation of evidence followed by judgment. The framework comprises a prosecution stream, a defense stream, and a judge model. We first build a dual-hypothesis segmentation architecture on a shared multi-scale encoder, in which the prosecution stream asserts manipulation and the defense stream asserts authenticity. Guided by edge priors, it produces evidence for manipulated and authentic regions through cascaded multi-level fusion, bidirectional disagreement suppression, and dynamic debate refinement. We further develop a reinforcement learning judge model that performs strategic re-inference and refinement on uncertain regions, yielding a manipulated-region mask. The judge model is trained with advantage-based rewards and a soft-IoU objective, and reliability is calibrated via entropy and cross-hypothesis consistency. Experimental results show that our model achieves superior average performance compared with SOTA IML methods.
