WW&&MM SScchhoollaarrWWoorrkkss Dissertations, Theses, and Masters Projects Theses, Dissertations, & Master Projects Fall 2016 TThhee IImmppaacctt ooff SSuurrffaaccee NNoorrmmaallss oonn AAppppeeaarraannccee Kathleen Dorothy Moore College of William and Mary, [email protected] Follow this and additional works at: https://scholarworks.wm.edu/etd Part of the Computer Sciences Commons RReeccoommmmeennddeedd CCiittaattiioonn Moore, Kathleen Dorothy, "The Impact of Surface Normals on Appearance" (2016). Dissertations, Theses, and Masters Projects. Paper 1477068344. http://doi.org/10.21220/S2Z30S This Dissertation is brought to you for free and open access by the Theses, Dissertations, & Master Projects at W&M ScholarWorks. It has been accepted for inclusion in Dissertations, Theses, and Masters Projects by an authorized administrator of W&M ScholarWorks. For more information, please contact [email protected]. The Impact of Surface Normals on Appearance Kathleen Dorothy Moore Hanson, Massachusetts Bachelor of Arts, College of the Holy Cross, 2009 Master of Science, College of William and Mary, 2011 A Dissertation presented to the Graduate Faculty of the College of William and Mary in Candidacy for the Degree of Doctor of Philosophy Department of Computer Science The College of William and Mary August 2016 © Copyright by Kathleen Dorothy Moore 2016 ABSTRACT The appearance of an object is the result of complex light interaction with the object. Beyond the basic interplay between incident light and the object's material, a multitude of physical events occur between this illumination and the microgeometry at the point of incidence, and also beneath the surface. A given object, made as smooth and opaque as possible, will have a completely different appearance if either one of these attributes - amount of surface mesostructure (small-scale surface orientation) or translucency - is altered. Indeed, while they are not always readily perceptible, the small-scale features of an object are as important to its appearance as its material properties. Moreover, surface mesostructure and translucency are inextricably linked in an overall effect on appearance. In this dissertation, we present several studies examining the importance of surface mesostructure (small-scale surface orientation) and translucency on an object's appearance. First, we present an empirical study that establishes how poorly a mesostructure estimation technique can perform when translucent objects are used as input. We investigate the two major factors in determining an object's translucency: mean free path and scattering albedo. We exhaustively vary the settings of these parameters within realistic bounds, examining the subsequent blurring effect on the output of a common shape estimation technique, photometric stereo. Based on our findings, we identify a dramatic effect that the input of a translucent material has on the quality of the resultant estimated mesostructure. In the next project, we discuss an optimization technique for both refining estimated surface orientation of translucent objects and determining the reflectance characteristics of the underlying material. For a globally planar object, we use simulation and real measurements to show that the blurring effect on normals that was observed in the previous study can be recovered. The key to this is the observation that the normalization factor for recovered normals is proportional to the error on the accuracy of the blur kernel created from estimated translucency parameters. Finally, we frame the study of the impact of surface normals in a practical, image-based context. We discuss our low-overhead, editing tool for natural images that enables the user to edit surface mesostructure while the system automatically updates the appearance in the natural image. Because a single photograph captures an instant of the incredibly complex interaction of light and an object, there is a wealth of information to extract from a photograph. Given a photograph of an object in natural lighting, we allow mesostructure edits and infer any missing reflectance information in a realistically plausible way. TABLE OF CONTENTS Acknowledgments iii Dedication iv List of Tables v List of Figures vi 1 Introduction 2 2 Background 5 2.1 Bidirectional Reflectance Distribution Function (BRDF) . . . . . . . 5 2.2 Types of BRDFs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.3 Image Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2.4 Bidirectional Subsurface Scattering Reflectance Distribution Func- tion (BSSRDF) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 3 Impact of Translucency on Normal Estimation 21 3.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 3.3 Prior work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.4 Empirical Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 4 Normal Estimation Refinement 42 4.1 Prior Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 i 4.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 4.3 Project Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.5 Discussion & Limitations . . . . . . . . . . . . . . . . . . . . . . . . 53 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 5 Image-Based Microgeometry Manipulation 56 5.1 Prior Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 5.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59 5.3 System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 5.5 Discussion & Limitations . . . . . . . . . . . . . . . . . . . . . . . . 72 5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 6 Conclusion 78 ii ACKNOWLEDGMENTS Pieter, thank you for taking me on and guiding me through the multi-dimensional world of computer graphics. Your accessibility will go unmatched, your mastery of the field - supreme, and your love of good teaching will be my benchmark, whether or not I end up teaching. I am grateful to have this lifelong connection in learning with you. Thank you, also, to my committee. I have great respect for your experience and knowledge and am honored to have my work vetted by you. I would like to offer special thanks to Dr. Bo Dong, my research partner and good friend. I have missed you in my last year at the College. Thank you for diving into problems with me and laughing whenever possible. To the rest of the graphics group: thank you for being a friendly and supportive bunch. Keep going, study hard. This work was supported in part by the U.S. National Science Foundation under grants IIS-1350323 and IIS-1016703. iii Even when I missed out on so much, working on my degree 454 miles away, my family supported me in numerous ways. I would like to dedicate this dissertation to them - especially to my Dad, who pressed me with questions on a regular basis and who is now finding answers, along with me, in my heart. iv LIST OF TABLES 4.1 BSSRDFparametersestimatedusingthejointmulti-channeldecon- volution versus ground truth parameters. . . . . . . . . . . . . . . . 55 v