Shape reconstruction from two color images using photometric stereo combined with segmentation and stereopsis

Most objects in a scene of our interest in reconstructing their shapes may have colors on them. Existing photometric methods require at least three images with different light directions to reconstruct the shape. If accurate shapes are obtainable even from two images, the construction of such an observation system may be much easier and its operation may be much speedier. In this paper we combine our photometric stereo method with color segmentation and the binocular stereopsis. First, color segmentation is applied to the images and component shapes are reconstructed from the segmented images. Then, the feature points of the components are estimated using the stereopsis not only to put them together to get the whole shape but also to reconstruct the scene. Basic experimental results are presented.