Communication of Digital Material Appearance Based on Human Perception
Abstract
In daily life, we encounter digital materials and interact with them in numerous situations, for instance when we play computer games, watch a movie, see billboard in the metro station or buy new clothes online. While some of these virtual materials are given by computational models that describe the appearance of a particular surface based on its material and the illumination conditions, some others are presented as simple digital photographs of real materials, as is usually the case for material samples from online retailing stores. The utilization of computer-generated materials entails significant advantages over plain images as they allow realistic experiences in virtual scenarios, cooperative product design, advertising in prototype phase or exhibition of furniture and wearables in specific environments. However, even though exceptional material reproduction quality has been achieved in the domain of computer graphics, current technology is still far away from highly accurate photo-realistic virtual material reproductions for the wide range of existing categories and, for this reason, many material catalogs still use pictures or even physical material samples to illustrate their collections. An important reason for this gap between digital and real material appearance is that the connections between physical material characteristics and the visual quality perceived by humans are far from well-understood. Our investigations intend to shed some light in this direction. Concretely, we explore the ability of state-of-the-art digital material models in communicating physical and subjective material qualities, observing that part of the tactile/haptic information (eg thickness, hardness) is missing due to the geometric abstractions intrinsic to the model. Consequently, in order to account for the information deteriorated during the digitization process, we investigate the interplay between different sensing modalities (vision and hearing) and discover that particular sound cues, in combination with visual information, facilitate the estimation of such tactile material qualities. One of the shortcomings when studying material appearance is the lack of perceptually-derived metrics able to answer questions like "are materials A and B more similar than C and D?", which arise in many computer graphics applications. In the absence of such metrics, our studies compare different appearance models in terms of how capable are they to depict/transmit a collection of meaningful perceptual qualities. To address this problem, we introduce a methodology to compute the perceived pairwise similarity between textures from material samples that makes use of patch-based texture synthesis algorithms and is inspired on the notion of Just-Noticeable Differences. Our technique is able to overcome some of the issues posed by previous texture similarity collection methods and produces meaningful distances between samples. In summary, with the contents presented in this thesis we are able to delve deeply in how humans perceive digital and real materials through different senses, acquire a better understanding of texture similarity by developing a perceptually-based metric and provide a groundwork for further investigations in the perception of digital materials.
Download: https://nbn-resolving.org/urn:nbn:de:hbz:5n-55049
Bibtex
@PHDTHESIS{martin-2019-dissertation, author = {Mart{\'i}n, Rodrigo}, title = {Communication of Digital Material Appearance Based on Human Perception}, type = {Dissertation}, year = {2019}, month = apr, school = {University of Bonn}, abstract = {In daily life, we encounter digital materials and interact with them in numerous situations, for instance when we play computer games, watch a movie, see billboard in the metro station or buy new clothes online. While some of these virtual materials are given by computational models that describe the appearance of a particular surface based on its material and the illumination conditions, some others are presented as simple digital photographs of real materials, as is usually the case for material samples from online retailing stores. The utilization of computer-generated materials entails significant advantages over plain images as they allow realistic experiences in virtual scenarios, cooperative product design, advertising in prototype phase or exhibition of furniture and wearables in specific environments. However, even though exceptional material reproduction quality has been achieved in the domain of computer graphics, current technology is still far away from highly accurate photo-realistic virtual material reproductions for the wide range of existing categories and, for this reason, many material catalogs still use pictures or even physical material samples to illustrate their collections. An important reason for this gap between digital and real material appearance is that the connections between physical material characteristics and the visual quality perceived by humans are far from well-understood. Our investigations intend to shed some light in this direction. Concretely, we explore the ability of state-of-the-art digital material models in communicating physical and subjective material qualities, observing that part of the tactile/haptic information (eg thickness, hardness) is missing due to the geometric abstractions intrinsic to the model. Consequently, in order to account for the information deteriorated during the digitization process, we investigate the interplay between different sensing modalities (vision and hearing) and discover that particular sound cues, in combination with visual information, facilitate the estimation of such tactile material qualities. One of the shortcomings when studying material appearance is the lack of perceptually-derived metrics able to answer questions like "are materials A and B more similar than C and D?", which arise in many computer graphics applications. In the absence of such metrics, our studies compare different appearance models in terms of how capable are they to depict/transmit a collection of meaningful perceptual qualities. To address this problem, we introduce a methodology to compute the perceived pairwise similarity between textures from material samples that makes use of patch-based texture synthesis algorithms and is inspired on the notion of Just-Noticeable Differences. Our technique is able to overcome some of the issues posed by previous texture similarity collection methods and produces meaningful distances between samples. In summary, with the contents presented in this thesis we are able to delve deeply in how humans perceive digital and real materials through different senses, acquire a better understanding of texture similarity by developing a perceptually-based metric and provide a groundwork for further investigations in the perception of digital materials.}, url = {https://nbn-resolving.org/urn:nbn:de:hbz:5n-55049} }