Date Awarded

Summer 2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.)

Department

Computer Science

Advisor

Pieter Peers

Committee Member

Weizhen Mao

Committee Member

Denys Poshyvanyk

Committee Member

Andreas Stathopoulos

Committee Member

Rance Necaise

Abstract

Bidirectional Reflectance Distribution Functions, (BRDFs), describe how light is reflected off of a material. BRDFs are captured so that the materials can be re-lit under new while maintaining accuracy. BRDF models can approximate the reflectance of a material, but are unable to accurately represent the full BRDF of the material. Acquisition setups for BRDFs trade accuracy for speed with the most accurate methods, gonioreflectometers, being the slowest. Image-based BRDF acquisition approaches range from using complicated controlled lighting setups to uncontrolled known lighting to assuming the lighting is unknown. We propose a data-driven method for recovering BRDFs under known, but uncontrolled lighting. This approach utilizes a dataset of 100 measured BRDFs to accurately reconstruct the BRDF from a single photograph. We model the BRDFs as Gaussian Mixture Models, (GMMs), and use an Expectation Maximization, (EM), approach to determine cluster membership. We apply this approach to captured data as well as synthetic. We continue this work by relaxing assumptions about either lighting, material, or geometry. This work was supported in part by NSF grant IIS-1350323 and gifts from Google, Activision, and Nvidia.

DOI

http://dx.doi.org/10.21220/s2-fwpk-zr81

Rights

© The Author

Share

COinS