Fast algorithms for automatic moire fringe analysis: application to noncontact measurements for quality control of industrial components

摘要

Abstract: Moire methods are optical methods that are based on theeffect of superposition of grating lines and have beenwidely used in the context of industrial applicationsfor shape analysis, for non-contact measurements, andfor quality control of industrial components. Inapplications the following computations: imagefiltering, fringe skeletonizing and fringe numberinghave to be performed for each test object, beforecomparison between the numerically reconstructed testobject shape and its CAD model. In order to reduce thecomputing time required by the preceding computations,the inverse moire technique has been introduced byHarthong. Instead of using a grating made of parallelstraight lines, the inverse moire technique uses apre-computed specific gratin, that is formed of curvedlines such that the moire pattern is composed ofparallel straight fringes if the test object shape isconformed to its CAD model. Defects are thencharacterized by a deformation and a curvature of theseparallel fringes. In this paper, we present examplesshowing that standard fringe extraction by automaticthresholding is not that easy. To overcome thisdifficulty, we propose a four stage processalgorithmical approach that allows fringe detection ininverse moire images with high sensitivity andspecificity. First we used the well-known imageprocessing technique called unsharp masking, to enhancemoire image and to emphasize low contrasted fringes.The second step is to extract bright fringes by imagesegmentation and constrained contour modeling. Afterdetection of these bright fringes inside the zone ofinterest of the moire image, we get the thick skeletonof adjacent background and of dark fringes. The thirdstep is to skeletonize this thick skeleton of adjacentbackground and of dark fringes, using morphologicalthinning of well-composed sets, that assures that eachfringe skeleton will be one pixel thick, at thedifference of standard thinning techniques. The fourthstep is to apply a graph technique to isolate theindividual dark fringes. When all these four steps havebeen followed, one is left with a binary image showingthe dark fringe pattern skeleton. The experimentalresults that have been obtained have shown therobustness of this algorithmical approach, for theanalysis of noisy inverse moire images. !14