A robust image processing algorithm has been developed for detection of small and low contrasted defects, adapted to X-ray images of castings having a non-uniform background. The sensitivity to small defects is obtained at the expense of a high false alarm rate. We present in this paper a feature extraction approach to complement the image processing, reducing the false alarms rate, while keeping a high defect detection rate, which is impossible by image processing techniques alone. ROC curves show a very good performance by using a new feature parameter, called 'Defect Confidence Index', combining three parameters and taking into account the fact that X-ray grey-levels follow a statistical normal law. Results are shown on a set of 684 images, involving 59 defects, on which we obtained a 100% detection rate without any false alarm.
|Journal||Insight - Non-Destructive Testing and Condition Monitoring|
|Publication status||Published - 2007|
Lecomte, G., Kaftandjian, V., Cendre, E., & Babot, D. (2007). A robust segmentation approach based on analysis of features for defect detection in X-ray images of aluminium castings. Insight - Non-Destructive Testing and Condition Monitoring, 49(10), 572-577. https://doi.org/10.1784/insi.2007.49.10.572