Methods of stochastic geometry in recognition of weld defects

Stochastic geometry is a valuable tool for fighting fuzziness inherent in some applied problems of pattern recognition. Defects in welds are hard to recognize because of the variability of their shapes, intensities, and background noise. To overcome these difficulties, a theory of feature recognition based on stochastic geometry was developed. In this paper, a new method for nonlinear filtering of images based on trace transform is suggested; it can be used to reduce noise, quantize images, and construct their polygonal approximations. A new class of features characterizing the weld shape, structure, and geometry is constructed. The efficiency of the method is confirmed by the results of experimental tests of a system recognizing weld defects.