Depth-Guided Dual-Domain Progressive Low-Light Enhancement for Light Field Image

In low-light environments, light field (LF) images are often affected by various degradation factors, which impair the performance of subsequent visual tasks such as depth estimation. To address these challenges, although numerous light-field low-light enhancement methods have been proposed, they generally overlook the importance of frequency-domain information in modeling light field features, thereby limiting their noise suppression capabilities. Moreover, these enhancement methods mainly rely on pixel- or semantic-level cues without explicitly incorporating disparity information for structural modeling, thereby overlooking the stereoscopic spatial structure of light field images and limiting enhancement performance across different depth levels. To address these issues, we propose a light field low-light enhancement method named DDPNet. The method integrates a depth-guided mechanism to jointly restore light field images in both the spatial and frequency domains, employing a multi-stage progressive strategy to achieve synergistic improvements in illumination and depth. Specifically, we introduce a Dual-Domain Feature Extraction (DDFE) module, which incorporates spatial-frequency analysis to efficiently extract both global and local light field features. In addition, we propose a Depth-Aware Enhancement (DAE) module, which utilizes depth maps to guide the enhancement process, effectively restoring edge structures and luminance information. Extensive experimental results demonstrate that DDPNet significantly outperforms existing methods.