Morphological Detection and Feature-Based Classification of Cracked Regions in Ferrites
by
Leszek Chmielewski
(1) Dept. Fund. Res. in EE, PAS, (2) Inst. Fund. Techn. Res., PAS, (3) Inst. Biocybernetics a. Biomed. Eng., (4) Association for Image Processing
Coauthors: Mariusz Nieniewski (1, 4), Adam Józwik (3), Marek Sklodowski (2, 4)

Automatic quality inspection of ferrite cores is a challenging task. The main difficulties in defect detection stem from the fact that the surface of ferrite cores is relatively dark and in many cases it is covered by a pattern of more or less regular stripes which represent traces of the grinding of some of the walls of the cores. This means that one deals with a very low signal-to-noise ratio. A 2-stage vision system for detection and measurement of crack regions was devised. In the first, detection stage the regions with strong evidence for cracks are detected. The main tool used for crack detection are morphological operations detecting irregular changes of brightness in the image. Subsequently a morphological reconstruction of cracks is carried out. The cracks usually consist of randomly distributed straight line segments, which are detected using the structuring elements in the form of short line segments directed at certain angles to the coordinate axis. Changing the thresholds for the binarization of the gray level map of cracks one obtains the marker and the mask necessary for the reconstruction of the binary map of cracks. The resulting map usually contains most of the information concerning the cracks together with some undesired information on the stripes coming from the grinding. The traces of grinding represent the false positive classification error. The second stage of the described vision system includes a feature-based K-nearest neighbor fuzzy classifier, which analyzes all the pixels indicated by the reconstructed binary map of cracks. The classifier uses reference patterns generated by a preparatory process including reclassification and replacement procedures. The combination of the morphological detector with the K-nearest neighbor classifier gives more precise results in a reasonable time. The detector is fast, but it assigns too many pixels to cracks. The classifier checks the results of the detector and in general reduces the number of pixels assigned to cracks. However, the classifier is much slower, hence the detector is indispensable for reduction of the amount of data processed by the classifier. Experimental results obtained with the described vision system are presented.

http://www.ippt.gov.pl/~kies/spokomm.htm

Date received: February 12, 1999

Copyright © 1999 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cacp-44.