Glare Removal for Astronomical Images with High Local Dynamic Range

Veiling glare is a common imaging corruption, describing the phenomenon of light spreading from bright regions into otherwise dark regions. This leads to a reduction of contrast, locally limiting high dynamic range imaging capabilities in addition to what the sensor’s pixels are able to record. In astronomical imaging, which motivates the present study, high local dynamic ranges are ubiquitous, an extreme example being the contrast between the moon and the sun’s corona during a total solar eclipse. While this corruption phenomenon has been known for a long time, the field lacks a specialized method for this problem. Building on the structure of the glare-corruption operator, we devise an efficient framework which can be combined with off-the-shelf optimization methods to address the problem of glare removal. We also present a novel evaluation set consisting of pairs of images with and without glare. We show state-of-the art glare removal performance on this benchmark and other real world images, including telescope images of a total solar eclipse.