Shape-from-shading(SFS) is one of the important approaches of 3-D surface reconstruction in computer vision. Since reflectance map equation in SFS is a nonlinear partial differential equation(PDE) with two unknown var...Shape-from-shading(SFS) is one of the important approaches of 3-D surface reconstruction in computer vision. Since reflectance map equation in SFS is a nonlinear partial differential equation(PDE) with two unknown variables, SFS with one image is ill-posed in mathematical sense. A linear perspective SFS method with two images is proposed to deal with the problem. We assume that two images with different light source directions are captured firstly. Orthogonal projection is not as accurate as perspective one to simulate imaging processes. Two reflectance map equations are established based on the Lambertian model under perspective projection, and the equations are further transformed into one linear PDE. Then the iterative semi-Lagrangian algorithm is used to approximate the solution. Finally, 3-D height values of pixel points in imaging planes are solved by the numerical interpolation method. Experimental results of both hemisphere and complex surfaces show that the proposed method can reconstruct surfaces accurately.展开更多
基金supported by the National Natural Science Foundation of China(61005015)the Third National Post-Doctoral Special Foundation of China(201003280)
文摘Shape-from-shading(SFS) is one of the important approaches of 3-D surface reconstruction in computer vision. Since reflectance map equation in SFS is a nonlinear partial differential equation(PDE) with two unknown variables, SFS with one image is ill-posed in mathematical sense. A linear perspective SFS method with two images is proposed to deal with the problem. We assume that two images with different light source directions are captured firstly. Orthogonal projection is not as accurate as perspective one to simulate imaging processes. Two reflectance map equations are established based on the Lambertian model under perspective projection, and the equations are further transformed into one linear PDE. Then the iterative semi-Lagrangian algorithm is used to approximate the solution. Finally, 3-D height values of pixel points in imaging planes are solved by the numerical interpolation method. Experimental results of both hemisphere and complex surfaces show that the proposed method can reconstruct surfaces accurately.
