Computational Bounce Flash for Indoor Portraits

摘要

Portraits taken with direct flash look harsh and unflattering becausernthe light source comes from a small set of angles very close tornthe camera. Advanced photographers address this problem by usingrnbounce flash, a technique where the flash is directed towardsrnother surfaces in the room, creating a larger, virtual light sourcernthat can be cast from different directions to provide better shadingrnvariation for 3D modeling. However, finding the right directionrnto point a bounce flash requires skill and careful consideration ofrnthe available surfaces and subject configuration. Inspired by thernimpact of automation for exposure, focus and flash metering, wernautomate control of the flash direction for bounce illumination. Wernfirst identify criteria for evaluating flash directions, based on establishedrnphotography literature, and relate these criteria to the colorrnand geometry of a scene. We augment a camera with servomotorsrnto rotate the flash head, and additional sensors (a fisheye andrn3D sensors) to gather information about potential bounce surfaces.rnWe present a simple numerical optimization criterion that finds directionsrnfor the flash that consistently yield compelling illuminationrnand demonstrate the effectiveness of our various criteria in commonrnphotographic configurations.