Automatic fringe pattern recognition using invariant moments - a feasibility study

The various techniques of holographic non-destructive testing, including Speckle interferometry and shearography, have the common property to reflect material flaws as irregularities in some kind of fringes. Automatic testing requires a computerized recognition of typical fringe patterns corresponding to the defects. Methods of artificial intelligence seem to be promising, as for example neural networks or fuzzy classification. However, the main prerequisite for their application is to drastically reduce the amount of data representing the pattern. This paper will report on how to describe fringe patterns by invariant moments in the space and frequency domain, respectively. A similarity concept for fringe patterns is introduced and test are performed with sets of synthetic patterns. The influence of Speckle noise and reduced fringe contrast is discussed. Finally, the moments are used to define features for an automatic flaw classification by fuzzy logic algorithms.