Visual inspection of metal surfaces

The majoTity of applications of automatic visual inspection have been the case in which a high contrast image can be obtained. This will result from object silhouettes 1 and high contrast reflectivity changes as in printed text. In these cases, the image can usually be successfully segmented by a threshold operation2,3 leading to a two-level or binary image. The cases that lead to difficulty at present involve the use of reflected light. Here one is faced with shadows and highly variable reflected intensity. This is the case for metal surfaces where the reflected intensity is a strong function of illumination and viewing direction. On the other hand, the inspection of metal surfaces represents an important domain of applications. Of particular interest is the detection of small surface defects such as nicks and scratches. In order to implement automatic computer inspection of metal surfaces, optical and illumination means must be Plv vided that provide high contrast images of surface defects. In addition, there must be a relationship between image intensity and surface profile. This paper will discuss a number of theoretical aspects of the scattering of light from metal surfaces. This provides a basis for computer modeling of the metal surface. The theory is based on earlier work by Beckmann 4 and Horn. 5 The method has been tested experimentally and several examples will be demonstrated.