Shape reconstruction for color objects using segmentation and photometric stereo

Objects whose shapes we want to estimate may have not a single color but mostly multicolors. This may make it difficult to estimate their shapes using the existing photometric stereo methods in a straightforward manner. In this paper we combine segmentation with photometric stereo to reconstruct shapes for multicolor objects. For this purpose, our previous photometric stereo method that uses two images is extended to use three images, with modifications in the application of boundary conditions. The three images are segmented in color and the resulting segmented images are normalized in each region in intensity, based on the assumption that the three lighting directions are apart enough from each other. The modified shading images are used to reconstruct the object's shape. Experiments using synthesized images clearly show the usefulness of our method.