Photometric Object Modeling -- Rendering from a . . .

For some time, rendering photorealistic synthetic images from observations of real objects has been a major research topic in the computer vision and computer graphics communities. An extensive amount of work in this research area has lead to a few representative schools: image-based rendering, inverse rendering and 3D photography. However, each of these methods still suffers from several drawbacks, such as massive data storage, constraints on applicable objects, restrictive scenarios on application, etc. To overcome these problems, we propose twomethods. The first method, Eigen-Texture Rendering, handles the appearance variation of the target object on its own 2D surface, enabling effective compression and interpolation with PCA thereby resulting in a compact representation of objects with arbitrary reflectance properties. The second method, Rendering from a Sparse Set of Images, recovers both the illumination distribution and reflection parameters simultaneously from input images, providing even more compact representation for photorealistic image synthesis. With the leverage of assuming a specific reflection model and restricting the treatable objects, the latter method provides more flexibility in application, i.e., fewer input images. In this dissertation, we present the theory of these methods, and report on the results obtained by applying the methods to real world objects