查看文章

This thesis concerns the field of machine vision, in which computers are used to interpret pictures provided by television cameras. In particular it tackles the problems of interpreting pictures containing the glossy highlights produced by specular reflection. These specular highlights or specularities occur only where a glossy surface reflects the direct illumination of a light source almost straight into the camera's aperture-as if the surface were a mirror. Practical vision systems ought to detect specularities for two reasons: firstly so that they will not be mistaken as surface markings, and secondly so that they can be used to infer shape information about the surface upon which they appear to lie. Schemes for both detection and shape inference are discussed and developed in this thesis.

Surfaces which are not glossy-Lambertian surfaces-form images of a certain constrained class. The detection scheme described in this thesis identifies regions. of. an. image where the Lambertian constraints break. These regions are candidate specularities. In all, the evidence provided by tests on three separate constraints is combined to detect specularities. The scheme will work even for achromatic ('black and white') cameras.