Detecting Specular Reflections Using Lambertian Constraints

Specularities-bright image regions formed by specular reflection-are likely to be mistaken for genuine surface markings by processes that perform photometric analysis, derive motion fields from optical flow or est imate depth from binocular stereo. In addition th ey can be used to infer surface geometry. This paper describes a scheme for detecting specularities based on a characterisation of Lambertian surfaces. Real surfaces are not simple composites of Lambertian and specular surfaces but evidence suggests that away from speculariti es they do not deviate fromthe Lambertian model by more than a factor of 3 or so. Thisproves adequate to serve as a constraint for specularity detection. Two independent tests identify image regions where the constraint is violated. The detector is shown to performs well on a varie ty of real images. Applications: improving binocular stereo and inferr ing surface geometry are described using specularities produced by the detector.