Arbitrary 3D shape-from-shading with parametric eigenspace method using image scanner

This paper shows the shape reconstructing method from an image taken by an image scanner. The shape reconstruction from the images is the important problem in the field of computer vision. One of the method of this problem is "shape-front-shading" which recovers an object surface shape from the shading (brightness) in the image. This method usually uses the photometric models about the reflection on an object surface, and solve this shape reconstruction problem as an energy minimizing problem using the non-linear photometric models. Therefore, it takes much costs for computing. To overcome such problem, we show another shape-from-shading method using several samples of shading images and shapes. To reduce the size of sample shading and shape data, we restore these data to the eigenspace. We also show the shape reconstruction method using the eigenspace data. The basic idea is the linear interpolation between the shading and shape eigenspaces. In order to recover the shape precisely, we consider the noise factors of the interpolation. In the experiments, we show the effectiveness of the proposed method by recovering the real object shape from the scanned image.