Afifi Mahmoud ; Derpanis Konstantinos G. ; Ommer Björn ; Brown Michael S.

Abstract

Capturing photographs with wrong exposures remains a major source of errorsin camera-based imaging. Exposure problems are categorized as either: (i)overexposed, where the camera exposure was too long, resulting in bright andwashed-out image regions, or (ii) underexposed, where the exposure was tooshort, resulting in dark regions. Both under- and overexposure greatly reducethe contrast and visual appeal of an image. Prior work mainly focuses onunderexposed images or general image enhancement. In contrast, our proposedmethod targets both over- and underexposure errors in photographs. We formulatethe exposure correction problem as two main sub-problems: (i) color enhancementand (ii) detail enhancement. Accordingly, we propose a coarse-to-fine deepneural network (DNN) model, trainable in an end-to-end manner, that addresseseach sub-problem separately. A key aspect of our solution is a new dataset ofover 24,000 images exhibiting the broadest range of exposure values to datewith a corresponding properly exposed image. Our method achieves results on parwith existing state-of-the-art methods on underexposed images and yieldssignificant improvements for images suffering from overexposure errors.