Image-based rendering (IBR) is one of the most promising techniques for photorealistic image synthesis. However, the naive IBR framework has some drawbacks, including restrictions on viewing direction, and difficulties in managing illumination change or object deformation. In this paper, we present a new IBR method capable of composing images of deformable objects from arbitrary view points, and under arbitrary illumination. To do this, we measured and utilized geometric object model and bidirectional texure function (BTF). BTF is generalization of BRDF with spatial variation along object's surface, represented as a set of texture databases which are captured from every viewing angle and light direction. Th evaluate the efficiency of this method, we performed several experiments on objects with non-rigid characteristics (e.g., cloth with varying shininess, a 3D object with complicated surface attributes) which are difficult to render correctly by general model-based CG techniques. Compared to previous BRDF based rendering approaches, our work more fully utilizes the 4-dimensional lighting/viewing parameters of BTF; since it is essential for image composition of deformed objects. Further, our implementation sorts the data based on BRDF parameters, resulting in compact data representation.
展开▼