Abstract

The detection of defects in X-ray testing follows a pattern recognition scheme where feature extraction plays a very significant role. In this paper, we present a new feature based on the number of high-contrast pixels located inside a segmented potential defect related to the size of the potential defect. The developed feature can be easily computed and offers a high separability according to the Fischer linear discriminant. The feature depends on only two parameters that can be automatically determined in a training phase. The developed feature and other reported features are tested in 72 radioscopic images of aluminium wheels. The comparison shows that the separability of the developed feature is at least six times higher than the separability achieved by other features.