Tactile-Assisted Low-Light Image Enhancement

Visual images captured in extremely low-light conditions often suffer from invisibility, low contrast, and amplified noise. While current vision-only enhancement methods achieve limited improvements in controlled settings, they fail to perform effectively in unconstrained environments due to inherent photonic limitations. In this paper, we propose Vi-taLight, a novel tactile-assisted low-light image enhancement framework that pioneers a new paradigm for extreme lowlight scenarios, particularly targeting challenging applications like industrial inspection and robotics vision. The core innovation of VitaLight lies in the Texture Supplementation Module (TSM), a plug-and-play component that integrates highfidelity texture priors derived from tactile images through a gradient-magnitude-aware gating mechanism. This enables cross-scale tactile-visual matching and adaptive texture supplementation. With minimal computational overhead and seamless adaptability, TSM offers a practical solution for lowlight image enhancement. Extensive experiments show that models equipped with the TSM significantly outperform baseline models across multiple datasets and our TSM generalizes to different unsupervised and supervised models well. The source code will be made public after peer review process.