AMAS Workshop on Image Analysis in Investigation of Concrete SIAIC’02 – (pp.203-230) – Warsaw, October 21-23, 2002. Classification of defects on the surface of black ceramics ∗) L. CHMIELEWSKI, M. SK LODOWSKI and W. CUDNY Institute of Fundamental Technological Research ´ Swi¸ etokrzyska 21, 00-049 Warszawa, Poland e-mail: {lchmiel,msklod,wcudny}@ippt.gov.pl Methods of visual detection and measurement of certain surface defects of black ceramics – magnets and ferrite cores – are presented. Difficulties such as dark colour of ferrite and variability of appearance of defects under different light con- ditions have been overcome. The monochromatic images of the product surface, taken under a highly specialised lighting system with a standard monochroma- tic digital camera, are processed with a hierarchical morphological irregularity detector, classified with a feature-based k nearest neighbours classifier, and post- processed. The same methodology can be applied to other materials having similar appearance. Key words: quality inspection, surface defects, machine vision, pattern recognition, magnets, ferrite cores. 1. Introduction Ferrites are ceramic magnetic materials made of oxide powders. Ferrite cores are manufactured by compacting and sintering these powders, and by grinding some surfaces to required dimensions. The similar technology is applied in the production of other products made of black ceramics. The aim of the paper is a presentation of the methods of detection and measurement of certain surface irregularities which emerge in the manufacturing process of magnets and ferrite cores. The main difficulties encountered are: dark colour of ferrite, large variety of possible shapes and locations of defects, and variability of their appearance, even in typical light conditions [9]. The above circumstances have effectively prevented even the leading ma- nufacturers of ferrites from automating the surface quality inspection of fer- rite cores. Manufacturers and buyers of ferrites distinguish tens of classes of defects