It is shown that the 3-D shape of a real world object may be determined by illuminating it by polarized illumination and performing polarization analysis on light reflected by its surface. The normal vector to each surface point is determined by the intersection of two planes (the illuminating one and the viewing one). While one base vector of each constraining plane comes from the scene geometry (the illumination vector and the viewing vector, respectively), the other is constructed from evaluated polarization parameters.

Unlike conventional physics-based methods (the shape from shading or the photometric stereo, for example), our method does not rely on exact knowledge of the model of reflectance, but rather assumes a class of reflectance models which share the same directional properties (i.e. the behavior with respect to the light polarization). As a consequence, it is more robust to undesireable but commonly observed effects in images (like interreflections). From its nature, the method is capable to distinguish between the body and the interface reflections, and opens the way for physically valid reflection components separation.