<p>Image copy-move forgery localization and source/target differentiation can be formulated as a cooperative visual reasoning task that requires the integration of duplicated-region correspondence and manipulation-induced boundary anomalies. Motivated by the cognitive principles of functional specialization, hierarchical processing, and cooperative evidence integration, we propose a cognitively inspired peer-interaction network, named PINet. PINet contains two computational pathways with complementary perceptual roles. The similarity detection branch is interpreted as a homology-perception peer that captures intra-image duplicated-region correspondence, while the artifact detection branch is interpreted as a boundary-anomaly-perception peer that learns boundary-sensitive representations under edge-aware auxiliary supervision, which serves as a proxy for transformation-induced boundary anomalies. Their representations interact at corresponding semantic levels and are aggregated across scales before the final pixel-level prediction. This formulation represents a principle-level computational abstraction rather than a direct model of human neural circuitry. Experiments on four public copy-move forgery datasets show that PINet achieves competitive performance among the evaluated methods under the adopted protocols for both three-class source/target differentiation and binary copy-move localization. Ablation results indicate that edge-aware auxiliary supervision and the complete peer-interaction fusion module improve source/target prediction relative to the corresponding ablated variants. The current framework remains limited by sensitivity to heavy post-processing, dependence on dense annotations, and the additional computational cost introduced by the dual-branch architecture and self-correlation operation.</p>