DPCformer: Dual-Prior Calibration Transformer for low-light image enhancement

Low-light image enhancement (LLIE) remains challenging because low-light degradation usually exhibits complex and coupled characteristics, including uneven illumination, weakened structures, and amplified noise. The aim of this paper is to develop an effective and efficient LLIE framework for coordinated exposure correction, structure preservation, and noise-aware detail recovery under complex low-light conditions. To this end, this paper proposes a Dual-Prior Calibration Transformer (DPCformer), which jointly exploits illumination-aware and content-aware priors in a unified restoration framework. Specifically, a Dynamic Illumination Routing Estimator (DIRE) is designed to adaptively aggregate multiple illumination pathways for spatially varying illumination modeling. A Global–Local Feature Modulator (GLFM) is introduced to capture compact global context and frequency-aware local structures, thereby improving detail preservation while suppressing unreliable high-frequency responses. A Dual-Prior Calibration Attention (DPCA) module further integrates illumination and content guidance into the attention process, enabling selective enhancement of illumination-sensitive regions while preserving reliable structural information. Extensive experiments on eight widely used benchmark datasets demonstrate that DPCformer outperforms recent state-of-the-art methods in restoration quality while maintaining competitive computational efficiency. Evaluations on downstream object and face detection tasks further demonstrate that the enhanced images provide more reliable visual cues for high-level vision tasks. Overall, the results indicate that cooperative calibration of illumination and content priors effectively improves brightness correction, structural fidelity, perceptual quality, and practical usability in low-light environments.