Photometric Stereo Using Four Surface Normal Approximations and Optimal Normalization of Images

We present a new method of photometric stereo, in which four surface normal approximations are introduced and images are normalized appropriately using a criterion. First, the method, which is based on application of the Jacobi iterative method to the reflectance map equation, introduces the four surface normal approximations and combines the resulting 4M iterative relations as constraints, where M is the number of images, aiming at reducing characteristic distortions in the reconstructed shape that could appear when only a single approximation is used. Next, to cope with those cases where none of object parts are normal to the lights, we normalize the images using a simple criterion of how close the shape is reconstructed to the given boundary value. The new method is examined using synthesized and real images, where two optical polarization filters are used in a combined way to get rid of the specular reflection components in the images so that the Lambertian reflection model is applicable