We investigate two different systems performing automatic visual inspection. The first is the inspection of highly reflective aluminum sheets, used by the Danish company Bang & Olufsen, as a part of the exterior design and general appearance of their audio and video products. The second is the inspection of IBM hard disk read/write heads for defects during manufacturing.
We have surveyed visual inspection system design methods and presented available image processing hardware to perform high resolution image capture. We present general usable practical visual inspection system solutions, when performing high resolution visual inspection of surfaces.
We have presented known and new lighting methods in a framework, general usable for inspecting reflective surfaces. Special attention has been given to the design of illumination techniques to enhance defects of highly reflective aluminum sheets. The chosen optical system setup has been used to enhance surface defects of other reflective surfaces, providing new and exciting applications subject to automated visual inspection.
Several contextual features have been surveyed along with introduction of novel methods to perform data-dependent enhancement of local surface appearance . Morphological methods have been described and utilized in algorithms for detecting 3-dimensional surface damages based on images from a novel structured lighting setup enhancing the appearance of these defects in specular surfaces.
A hardware implementable polynomial classifier structure has been described and compared to better known techniques based on multidimensional Gaussian models and tree classifiers. We have introduced a reject class definition for this classifier, and compared it against the classical Mahalanobis distance approach.
Finally, an evaluation of the total system performance in the case of inspecting reflective surface for scratches, and inspection of hard disk sliders is presented.