A Computational Approach for Obstruction-Free Photography

摘要

We present a unified computational approach for taking photosrnthrough reflecting or occluding elements such as windows andrnfences. Rather than capturing a single image, we instruct the user torntake a short image sequence while slightly moving the camera. Differencesrnthat often exist in the relative position of the backgroundrnand the obstructing elements from the camera allow us to separaternthem based on their motions, and to recover the desired backgroundrnscene as if the visual obstructions were not there. We show resultsrnon controlled experiments and many real and practical scenarios, includingrnshooting through reflections, fences, and raindrop-coveredrnwindows.