A deep thermal-guided approach for effective low-light visible image enhancement

•A new perspective on how to effectively extract thermal feature, emphasizing the extraction of edges/textures-related features in different receptive fields on thermal images.•A novel module to generate cross-modal dependent and spatially-variant kernels for the enhancement of weak low-light visible features under the guidance of the extracted thermal features.•The proposed model represents the first attempt to successfully perform low-light visible image enhancement by incorporating complementary information presented in the thermal channels. It outperforms state-of-the-art methods on both simulated and realistic nighttime images in terms of restoration accuracy, visual perception, and computational efficiency.