Abstract
This paper presents our computer vision research work to recover 3D shape and to render the virtual image based on the 3D shape. The approach can handle neural network to construct 3D model from the combination of shading images taken under different directions of light source. The shape parameters extracted in the approach are surface gradient and Gaussian curvature. Neural network can be used to generate a virtual image from any view point with any light source. It is shown how to construct neural network to recover the shape and to render the virtual image. The approach are empirical without any assumptions for surface reflectance or directions of light source, that is, non-parametric functional approximation can be achieved with neural network through the shape reconstruction and rendering processes.
