In this paper we present a software tool by which the image processing and defect classification parts of an inspection system can more easily be designed and tested. The user interface of the tool has been programmed using Microsoft Visual Basic to which the C coded development software has been liked as Windows DLL-libraries. At the moment, a texture analysis method based on sum and difference histograms has been implemented as a basic segmentation method, but the addition of other algorithms is quite straight-forward. In addition to texture analysis, the tool includes a set of classification functions for automatic generation of nearest neighbor and decision tree classifiers. Starting from raw image data, generation of an executable decision tree classifier is possible without writing a line of code. The generation of the decision tree classifiers is based on IFD3, which is an extended version of Quinlan's ID3 algorithm. In this extended version, more advanced features of the induction of fuzzy and hybrid decision trees have been introduced.
|Proceedings of SPIE
|European Symposium on Optics for Productivity in Manufacturing
|Frankfurt am Main
|20/06/94 → 24/06/94