Robust and efficient automated detection of tooling defects in polished stone

  • Authors:

  • J. R. J. Lee;M. L. Smith;L. N. Smith;P. S. Midha

  • Affiliations:

  • Machine Vision Laboratory, Faculty of Computing, Engineering and Mathematical Sciences, University of the West of England, Bristol, UK;Machine Vision Laboratory, Faculty of Computing, Engineering and Mathematical Sciences, University of the West of England, Bristol, UK;Machine Vision Laboratory, Faculty of Computing, Engineering and Mathematical Sciences, University of the West of England, Bristol, UK;Machine Vision Laboratory, Faculty of Computing, Engineering and Mathematical Sciences, University of the West of England, Bristol, UK

  • Venue:
  • Computers in Industry - Special issue: Machine vision
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

The automated detection of process-induced defects such as tooling marks is a common and important problem in machine vision. Such defects are often distinguishable from natural flaws and other features by their geometric form, for example their circularity or linearity. This paper discusses the automated inspection of polished stone, where process-induced defects present as circular arcs. This is a particularly demanding circle detection problem due to the large radii and disrupted form of the arcs, the complex nature of the stone surface, the presence of other natural flaws and the fact that each circle is represented by a relatively small proportion of its total boundary. Once detected and characterized, data relating to the defects may be used to adaptively control the polishing process. We discuss the hardware requirements of imaging such a surface and present a novel implementation of a randomised circle detection algorithm that is able to reliably detect these defects. The algorithm minimizes the number of iterations required, based on a failure probability specified by the user, thus providing optimum efficiency for a specified confidence whilst requiring no prior knowledge of the image. The probabilities of spurious results are also analyzed, and an optimization routine introduced to address the inaccuracies often associated with randomized techniques. Experimental results demonstrate the validity of this approach.