Deep Snapshot HDR Reconstruction Based on the Polarization Camera

Abstract

In this thesis, we propose the use of the polarization camera for high-dynamic-range (HDR) imaging. Specifically, observing that natural light can be attenuated differently by varying the orientation of the polarization filter, we treat the multiple images captured by the polarization camera as a set captured under different exposure times, to support the development of solutions for the HDR reconstruction problem. Most existing methods are developed for conventional camera images. However, polarization cameras capture images differently than conventional cameras. In this thesis, we propose two deep snapshot HDR reconstruction frameworks, that uses polarimetric cues available from the polarization camera. With our deep-learning based methods, the obtained polarimetric information enables us to regress the missing pixels in polarization images more effectively. We train and validate the methods on our collected polarization dataset. We demonstrate through experimental results that our approach can reconstruct visually pleasing HDR results, and performs favorably than state-of-the-art HDR reconstruction algorithms. The source code is publicly available on Github.