Appearance Capturing Devices and Methods for Photo-realistic Representation of Complicated 3D Objects

Abstract

The acquisition of real objects in a digital form has been widely explored in recent years along with advances in computer graphics technologies. To represent an object in a photo-realistic manner, the geometrical shape and appearance of the object must be considered. Therefore, accurate 3D reconstruction and realistic appearance capturing and modeling technologies have steadily developed in these fields. There has been a recent emergence of new techniques that can capture the appearance of a model in an exact and efficient manner. However, such techniques have yet to reach the actual application stage, such as for use in museums and exhibitions, owing to a trade-off between rendering quality and the data capture and processing time. To tackle this problem, we designed a new appearance capturing system that can efficiently capture sample images for appearance modeling. In addition, we proposed a novel method that relatively reduces the required appearance capturing time while maintaining the rendering quality.

In the first study, an appearance capturing system was designed for efficiently acquiring the appearance of real-world objects. To create an efficient measuring device, we produced a total of three appearance capturing devices. The first was developed to acquire the meso-structure of a 2D object. The second and third devices differ only in the type of light sources and angular resolution applied, whereas the other factors are the same. The third proposed measurement system comprises a camera arc with 11 cameras and a light arc with 17 light sources, allowing the full hemispherical range to be covered over the target object by rotating and capturing the sample images at the user-defined positions. This system enables us to overcome the limitations of a conventional brute-force appearance capturing system requiring a large number of cameras, light sources, and highly complex systems. The capturing devices apply a system design that minimizes the amount of equipment required. Camera and light calibration are also proposed for a precise capturing of the sample images for appearance modeling, including a bidirectional reflectance distribution function (BRDF), spatially varying BRDF (SVBRDF), and bidirectional texture function (BTF). The system can capture 54,043 sample images that can be used for appearance modeling when varying from 187 viewing directions and 289 illumination directions.

In the second study, we proposed a novel appearance capturing method for considerably decreasing the number of images required for SVBRDF modeling while maintaining the quality of the model achieved when applying several images. This method enables us to overcome the time consumed when applying a considerable number of object images to accurately define the SVBRDF modeling when complicated geometric components are exhibited. Along with this method, we presented an acquisition score and its corresponding objective function when considering the local geometry and the reflectance properties, allowing an estimation of the optimal viewing and illumination directions for the image acquisition, by which the object is measured using reduced number of images. The captured images are then processed to obtain an SVBRDF of the object. The experimental results demonstrate that the proposed method can measure the appearance of a 3D object with complicated shapes using a few hundred images and efficiently produce an SVBRDF of the object. In addition, the proposed method can render an object photo-realistically with fewer artifacts than existing methods when any arbitrary viewing and illumination directions are given.