Incorrect exposure in photography does more than brighten or darken an image—it also removes recoverable scene detail. A new paper on arXiv argues that fixing this problem is best done on RAW sensor data rather than on the already-processed sRGB output.
The authors point out that existing sRGB-based correction methods are constrained by the non-linear image signal processing pipeline that converts RAW data into display-ready images. RAW-based methods, they suggest, have typically been developed in a narrower way, though the provided abstract is cut off before the full contrast is given.
The proposed framework, named "Unified RAW Image Exposure Correction and Restoration via Linear Illumination and Frequency-Adaptive Enhancement," works directly on RAW data and uses frequency-adaptive enhancement to recover details. By operating in the linear illumination space, it aims to avoid the distortions introduced by the non-linear pipeline that limits sRGB approaches.