Reflection Removal Using Multiple Polarized Images with Different Exposure Times

Abstract

When we take a photograph through glass windows or doors, the foreground scene is reflected in the captured image. The reflected components overlap with the background scene and make object recognition and identification more difficult. This paper proposes a novel reflection removal method using multiple polarized images taken with different exposure times. To achieve a high accuracy reflection removal in high dynamic range scenes, in which photographed images have under-/over-exposed pixels, we introduce a minimization problem of weighted nonnegative matrix factorization (WNMF) with total variation regularization. To solve this minimization problem, we also introduce an alternating optimization scheme with the alternating direction method of multipliers (AO-ADMM). The advantages of the proposed method over some conventional methods are demonstrated in experiments of reflection removal using real-world images.