Shape description via shading images

摘要

As we perceive a world of three-dimensional objects, one of the most important property is that a smooth opaque object will give rise to a shade image. The shading can provide information for describing or recovering the shape of an object. A process performing this purpose is known as shape description from shading images. In this paper we propose a method called extended photometric stereo using four light sources to solve the shading problem. When the least square estimation method is employed, the shading problem happens if the reflectance map is not known exactly (except for the direction of light sources), and how the object reflects the light falling on it. In dealing with this process we have met the ill-posed problem that there are not sufficient data to describe the shape of the object. Therefore, we develop iterative algorithms in accordance with the concepts of the regularization method and incorporation of the physical constraints to overcome the ill-posed problem. These algorithms can converge very soon and avoid the clumsiness of initial value assignment, and they are robust against noise. Our approach has been applied to both ideal and noisy computer-synthesized images to confirm its correctness and feasibility, and its performance has also been studied with a set of real images.