This research is to investigate techniques for modeling objects with complex reflectance. Our aim is to develop methods for recovering both shape and reflectance properties of natural objects -- objects that due to their inherent complexity have proven elusive to traditional shape reconstruction techniques such as binocular stereo, structure from motion, photometric stereo, and even laser range finders. Understanding the interplay between shape and reflectance is critical to object modeling and is in many ways central to the field of computer vision. Yet this promising avenue of research has been overlooked in the past, either due to acquisition and storage limitations or, perhaps, due to a lack of appropriate techniques. Recently, the PI's have introduced and prototyped two novel methods for shape reconstruction and reflectance recovery that essentially exploit two neglected physical principles. These two methods, named Helmholtz stereopsis and light field reconstruction, are able to recover surface shape regardless of the object's reflectance (BRDF), and they offer numerous advantages over conventional reconstruction techniques. The proposed research includes new shape reconstruction techniques, methods for constructing object models including their shape and BRDF on a point-by-point basis, and applications of these models to image-based rendering. In addition, these reconstruction techniques will serve as tools for a broad study of BRDF's in general as well as 3-D textures whose appearance varies with viewpoint and lighting.
Unlike current shape estimation techniques, the proposed methods will produce object models with both geometric and BRDF data. This new capability in object modeling technology will be beneficial in areas such as industrial design, architecture, entertainment, robotics, and medicine. Estimation of the reflectance function across the reconstructed objects, surface normals, and shape, leads directly to new methods for image-based modeling and rendering for computer graphics applications. These modeling techniques will be used to study reflectance properties of 3-D textures and natural materials (including human skin), and this will help to produce more effective generative appearance models that can be used in security applications (e.g., face recognition), human computer interaction applications, entertainment, learning through visual simulation, communications, etc. In addition to the applications of the resulting research and its dissemination through peer-reviewed publications, funding will be used to support the training of two Ph.D. students over three years.
