Foundations of computational photography and videography

点击翻译按钮获取中文摘要

Over the last decade, computational photography has emerged as a research field that promises to revolutionize the way we capture and interact with photos and videos. With digital cameras becoming ubiquitous and turning into full-fledged computing devices, it is now possible to interject significant computation before, during and after image acquisition. This has led to a fairly radical re-thinking of the camera's imaging chain---from the optics that transform incident light before it reaches the sensor, to the control algorithms operating the camera, to the algorithms that process its photos and videos afterward. Although the field has now gained a lot of momentum, long-term progress requires tackling fundamental imaging and analysis problems that are still not well understood. In this proposal I focus on three of these problems: (1) analyzing light transport, ie., how light flows through the scene being imaged; (2) achieving multi-focal photography, ie. designing a camera that can take many differently-focused photos at once; and (3) inferring the time-varying 3D shape of people, objects and environments appearing in a video or personal photo collection. While our work will be guided by computational photography applications, the problems themselves represent a natural continuation of core research in computer vision and computer graphics that I have been exploring with my group.