AGCSNet: High-contrast image-exposure correction with automatic illumination-map attention-based gamma and saturation correction

Abstract

Low-light image enhancement has made significant advancements in recent years. However, enhancing high-contrast images that exhibit both under- and overexposure remains a major challenge. To address this issue, we propose an exposure-correction method called AGCSNet. Two gamma corrections, ${\gamma }_1$ and ${\gamma }_2$, were applied separately to correct for underexposure and overexposure, producing two gamma-corrected images. An illumination map was used to differentiate between the underexposed and overexposed regions in the gamma-corrected images. To mitigate the saturation anomalies caused by the gamma corrections, we introduced a new saturation-correction method with a factor $s$. Moreover, optimal values for ${\gamma }_1,{\gamma }_2$, and $s$ can be predicted using our factor-estimation deep-learning model. We evaluated our method on eight datasets. In comparison with over 20 prior methods, our method demonstrates competitive performance with and, in some cases, surpasses state-of-the-art methods that are closely aligned with human visual perception.