Recovering Intrinsic Scene Characteristics from Images

摘要

The central theme of the research effort described in this report is the recovery of information about the three-dimensional structure and physical characteristics of the surfaces depicted in an image -- their shapes, locations, and photometric properties. Our approach has been to identify and model physically meaningful information that can either directly recover physical structure (e.g., models relating surface geometry to intensity variations visible in the image), or that can be used to constrain the interpretation process (e.g., camera and illumination models). In this report, we present new results concerning the question of how to recover the shape of a surface from its photometric variation in an image and a new method for skyline delineation. An essential feature of these techniques is that they are based entirely on physical models, and thus require no prior knowledge of the scene being analyzed.