Light field applications to 3-dimensional surface imaging

[摘要] The structure of light around a scene may be contained in a 4-dimensional array known as a light field. This thesis describes methods for acquiring and manipulating light fields for applications in 3-dimensional imaging. By actively sampling parts of the wavefront impinging on a lens, or using microlens arrays and patterned sinusoidal masks to modulate the rays reaching a camera, both the spatial distribution and directionality of light may be captured to produce light fields. Simple depth estimation algorithms using stereo and focus measures are then applied to recover quantitative depth information. Experiments on real-world light fields demonstrate their utility in performing digital refocusing, reconstructing occluded objects as well as accurately estimating depth and shape. The performance of the algorithms developed are discussed theoretically and compared empirically.

[发布机构] Massachusetts Institute of Technology