RI: Small: Higher Order Statistics for Appearance

University Of Southern California, Los Angeles CA

Investigators

Abstract

This project endeavors to develop novel computational representations for appearance modeling of physical materials for realistic computer graphics and machine vision applications by investigating spherical moments for modeling material appearance. Beyond the development of a novel theoretical framework for appearance modeling, the project involves evaluation of the generalization of the proposed appearance representation in various settings using software simulations as well using ground truth data available from various measurements. The project identifies three fundamental goals for such an appearance representation: application to general forward (reflectance modeling) and inverse (reflectance estimation) problems, appearance modeling in uncontrolled (lighting and viewpoint) settings, and validation and appearance classification for scene analysis applications. In particular, the analysis of higher order statistics for novel compact representations for forward simulation and inverse rendering problems are investigated. Additionally, appearance modeling from sparse input data acquired under general conditions of semi-controlled or uncontrolled lighting is investigated within this framework. Also investigated is the classification of material appearance from sparse measurements based on such spherical statistics. This research has far-reaching impact beyond computer graphics in many fields such as architecture, engineering, science, fine art and entertainment. Besides providing new insights into appearance modeling, the developed theory allows rapid measurements with greater ease, making the results more accessible to other researchers and practitioners in the field. Besides mentoring a graduate student, the findings of the research are planned to be integrated into a graduate level computer science course offering and through the creation of internship opportunities for undergraduate students.