Polarization analysis has already been applied to diverse problems in computer vision. Koshikawa (1979), Wolff (1987) and Koshikawa and Shirai (1987) recognised the potential of ellipsometric techniques to determine the orientation of dielectric surfaces, effectively a “shape from X” method. For example, Koshikawa and Shirai used several, circularly polarized light sources, causing specular reflections on the object surface so that the normals could be extracted at these points if the material properties were known. Using a facetted approximation, this was used to match scene against
known model facets to recognise and constrain the pose of the object (Koshikawa and Shirai, 1987). Jones and Fairney (1989) performed a similar task but using unpolarized, rather than circularly polarized light as the illumination source.
For precise instrumentation we have method yielded good pose estimates. These methods require very precise knowledge of the polarization state of both the incident and the reflected light, the latter in particular requiring accurate measurement of reflected intensity at different settings of the polarizing optics within the receiver. Knowledge of the material refractive index is also required. As Wolff remarked subsequently (Wolff, 1990), there are more effective methods for acquiring depth data.
Muller (1996) and others (Fryer and Miller, 1991; Mersch, 1984) have used polarization to distinguish
specular from diffuse reflections in a computer vision system. The crucial assumption in their work is that the specular reflections are unwanted and have higher intensity than the diffuse reflections. Muller used three orientations of a polarization filter to acquire three images, from which the specular and diffuse components were calculated using the Fresnel equations. Further simplifying assumptions about the scene and optical geometry and the material constants can eliminate the necessity to acquire three images; a single image acquired with a linear polarizer is sufficient. Nayar and coworkers (Nayar et al., 1997) observed that by considering colour and polarization simultaneously, more robust results could be achieved to discriminate between specular and diffuse reflections from a dielectric. This follows earlier work on the labelling of edges by examining the polarization of the light reflected from
adjacent dielectric surfaces (Boult and Wolff, 1991) and an even earlier description on the use of polarization to distinguish specular from diffuse reflection (Wolff, 1989). More recently, Chen and Wolff (1998) have considered that metal-dielectric discrimination is possible because metals and dielectrics differ in the phase change of polarized light on reflection.
source : Improving Depth Image Acquisition Using Polarized Light, A.M. WALLACE, B. LIANG, E. TRUCCO AND J. CLARK, 1997