Study On Image-based Modeling Techniques For Reconstruction Of Shape And Reflectance Properties Of Specular Objects

Image-based modeling is a fundamental technique in computer graphics and computer vision domain, which aims at recovering geometry of the scene. In constrast with 3D laser scan currently as a standard origin for shape, image-base modeling possesses particular advantages: economical, convenient and applicable. The popularization of digital camera makes these advantages more prominent. By image-based modeling, both shape and surface texture can be acquired for realistic rendering. This technique is widely used in many applications such as visualization, computer animation, object recognition, virtual reality, games and motion pictures.The elementary information, i.e., pixel intensity, is formed by radiance of a scene point. In physics, radiance is the interactional result of the surface material with the environmental lights. Besides diffuse reflection, specular reflection is also a common optical phenomenon for most real-world objects. The specular highlight in an image plays an important role in sensing the object shape, quality and lighting environment, so it does the same to realistic rendering. However, existence of specularities brings puzzles to shape recovery. In this dissertation, we have done a series works on shape recovery and reflectance property reconstruction for specular objects, including:1. Proposing an efficient shape recovery scheme by using HDR images. An accurate reflectance map under a specified illumination direction is acquired by radiance linear interpolation, and this map is applied in Shape from Shading as well as Photometric Stereo. Because diffuse and specular reflection are uniformly handled by using the radiance concept in physics sense, the reflectance map can replace any analytic BRDF model.2. Proposing an improve Photometric Stereo algorithm. Any possible specular reflection is represented as a linear combination of a group of specular terms with preset roughness values. With this representation, the specular object shape can be recovered, and furthermore, surface reflectance properties can be also obtained.3. Proposing a method for extracting realistic textures from real objects. A material weight map is formed by BRDF model parameters of each kind of"virtual"material and their composition scales. This map is texure-mapped to 3D models for realistic rendering with the help of BRDF parameters, and therefore, can be considered as a kind of"relighting"texture.4. Proposing a visual-hull-based method for recovering complete object shape and its reflectance properties. Multiple images of an object rotating around a fixed axis are used to extract all silhouettes for reconstruction and to calculate reflectance parameters. This is a systematic methodology for recovering both the completed surface shape and its reflectance properties.