Sasaki, Takahiro2021-04-122021-04-122020-01https://hdl.handle.net/11299/219298University of Minnesota Ph.D. dissertation. January 2020. Major: Scientific Computation. Advisor: James Leger. 1 computer file (PDF); x, 104 pages.We investigate the utility of plenoptic data for extracting information from a scene where the light from unknown objects in the scene is viewable only after scattering from a diffuse surface. A primary goal of this research is to estimate the objects’ locations in the hidden scene, and to extract additional information, such as the object’s shape and brightness. In the first part, we derive a rigorous relationship between the object and the scattered light fields, which is cast in terms of a system of Fredholm integral equations of the first kind with the BRDF of the scattering surface, and the object information is reconstructed by solving these equations. Based on the Fourier transformation, we propose a simple BRDF model and analyze the reconstruction errors by introducing newly defined parameters reflecting the BRDF’s characteristic, the degree of specularity and the effective SNR. We then obtain optimal regularized solutions to the equations under a variety of conditions. Moreover, we provide a fundamental limit of retrievable information content from the scattered light. A comparison with experimental results is reported. In the second part, we investigate the use of plenoptic data for locating objects from a scattered light field by using the results obtained in the first part. The resolution limits of the transverse and longitudinal location estimates are derived from fundamental considerations on the scattering physics and measurement noise. Based on the refocusing algorithm developed in the computer vision field, we derive a simple alternative formulation of the projection slice theorem in a form directly connecting the light field and spatial frequency spectrum. Using this alternative formulation, we propose a spatial frequency filtering method that is defined on a newly introduced mixed space-frequency plane and achieves the theoretically-limited depth resolution. Moreover, we propose an improved refocusing algorithm to more accurately estimate the object’s brightness information. An experimental verification is provided.enBRDFdepthinverse problemlight fieldrefocusingspatial frequency filteringThesis or Dissertation